Tag: nand flash

Posted on

Where, How to use NVMe overview primer

server storage I/O trends
Updated 1/12/2018

This is the fourth in a five-part miniseries providing a primer and overview of NVMe. View companion posts and more material at www.thenvmeplace.com.

Where and how to use NVMe

As mentioned and shown in the second post of this series, initially, NVMe is being deployed inside servers as “ back-end,” fast, low latency storage using PCIe Add-In-Cards (AIC) and flash drives. Similar to SAS NVM SSDs and HDDs that support dual-paths, NVMe has a primary path and an alternate path. If one path fails, traffic keeps flowing without causing slowdowns. This feature is an advantage to those already familiar with the dual-path capabilities of SAS, enabling them to design and configure resilient solutions.

NVMe devices including NVM flash AIC flash will also find their way into storage systems and appliances as back-end storage, co-existing with SAS or SATA devices. Another emerging deployment configuration scenario is shared NVMe direct attached storage (DAS) with multiple server access via PCIe external storage with dual paths for resiliency.

Even though NVMe is a new protocol, it leverages existing skill sets. Anyone familiar with SAS/SCSI and AHCI/SATA storage devices will need little or no training to carry out and manage NVMe. Since NVMe-enabled storage appears to a host server or storage appliance as an LUN or volume, existing Windows, Linux and other OS or hypervisors tools can be used. On Windows, such as,  other than going to the device manager to see what the device is and what controller it is attached to, it is no different from installing and using any other storage device. The experience on Linux is similar, particularly when using in-the-box drivers that ship with the OS. One minor Linux difference of note is that instead of seeing a /dev/sda device as an example, you might see a device name like /dev/nvme0n1 or /dev/nvme0n1p1 (with a partition).

Keep in mind that NVMe like SAS can be used as a “back-end” access from servers (or storage systems) to a storage device or system. For example JBOD SSD drives (e.g. 8639), PCIe AiC or M.2 devices. NVMe can also like SAS be used as a “front-end” on storage systems or appliances in place of, or in addition to other access such as GbE based iSCSI, Fibre Channel, FCoE, InfiniBand, NAS or Object.

What this means is that NVMe can be implemented in a storage system or appliance on both the “front-end” e.g. server or host side as well as on the “back-end” e.g. device or drive side that is like SAS. Another similarity to SAS is that NVMe dual-pathing of devices, permitting system architects to design resiliency into their solutions. When the primary path fails, access to the storage device can be maintained with failover so that fast I/O operations can continue when using SAS and NVMe.

NVM connectivity options including NVMe
Various NVM NAND flash SSD devices and their connectivity including NVMe, M2, SATA and 12 Gbps SAS are shown in figure 6.

Various NVM SSD interfaces including NVMe and M2
Figure 6 Various NVM flash SSDs (Via StorageIO Labs)

Left in figure 6 is an NAND flash NVMe PCIe AiC, top center is a USB thumb drive that has been opened up showing an NAND die (chip), middle center is a mSATA card, bottom center is an M.2 card, next on the right is a 2.5” 6 Gbps SATA device, and far fright is a 12 Gbps SAS device. Note that an M.2 card can be either an SATA or NVMe device depending on its internal controller that determines which host or server protocol device driver to use.

The role of PCIe has evolved over the years as has its performance and packaging form factors. Also, to add in card (AiC) slots, PCIe form factors also include M.2 small form factor that replaces legacy mini-PCIe cards. Another form factor is M.2 (aka Next Generation Form Factor or NGFF) that like other devices, can be an NVMe, or SATA device.

NGFF also known as 8639 or possibly 8637 (figure 7) can be used to support SATA as well as NVMe depending on the card device installed and host server driver support. There are various M.2 NGFF form factors including 2230, 2242, 2260 and 2280. There are also M.2 to regular physical SATA converter or adapter cards that are available enabling M.2 devices to attach to legacy SAS/SATA RAID adapters or HBAs.

NVMe 8637 and 8639 interface backplane slotsNVMe 8637 and 8639 interface
Figure 7 PCIe NVMe 8639 Drive (Via StorageIO Labs)

On the left of figure 7 is a view towards the backplane of a storage enclosure in a server that supports SAS, SATA, and NVMe (e.g. 8639). On the right of figure 7 is the connector end of an 8639 NVM SSD showing addition pin connectors compared to an SAS or SATA device. Those extra pins give PCIe x4 connectivity to the NVMe devices. The 8639 drive connectors enable a device such as an NVM, or NAND flash SSD to share a common physical storage enclosure with SAS and SATA devices, including optional dual-pathing.

Where To Learn More

View additional NVMe, SSD, NVM, SCM, Data Infrastructure and related topics via the following links.

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

Be careful judging a device or component by its physical packaging or interface connection about what it is or is not. In figure 6.6 the device has SAS/SATA along with PCIe physical connections, yet it’s what’s inside (e.g. its controller) that determines if it is an SAS, SATA or NVMe enabled device. This also applies to HDDs and PCIe AiC devices, as well as I/O networking cards and adapters that may use common physical connectors, yet implement different protocols. For example, the SFF-8643 HD-Mini SAS internal connector is used for 12 Gbps SAS attachment as well as PCIe to devices such as 8630.

Depending on the type of device inserted, access can be via NVMe over PCIe x4, SAS (12 Gbps or 6Gb) or SATA. 8639 connector based enclosures have a physical connection with their backplanes to the individual drive connectors, as well as to PCIe, SAS, and SATA cards or connectors on the server motherboard or via PCIe riser slots.

While PCIe devices including AiC slot based, M.2 or 8639 can have common physical interfaces and lower level signaling, it’s the protocols, controllers, and drivers that determine how they get a software defined and used. Keep in mind that it’s not just the physical connector or interface that determines what a device is or how it is used, it’s also the protocol, command set, and controller and device drivers.

Continue reading about NVMe with Part V (Where to learn more, what this all means) in this five-part series, or jump to Part I, Part II or Part III.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

NVMe Need for Performance Speed Performance

server storage I/O trends
Updated 1/12/2018

This is the third in a five-part mini-series providing a primer and overview of NVMe. View companion posts and more material at www.thenvmeplace.com.

How fast is NVMe?

It depends! Generally speaking NVMe is fast!

However fast interfaces and protocols also need fast storage devices, adapters, drivers, servers, operating systems and hypervisors as well as applications that drive or benefit from the increased speed.

A server storage I/O example is in figure 5 where a 6 Gbps SATA NVM flash SSD (left) is shown with an NVMe 8639 (x4) drive that were directly attached to a server. The workload is 8 Kbyte sized random writes with 128 threads (workers) showing results for IOPs (solid bar) along with response time (dotted line). Not surprisingly the NVMe device has a lower response time and a higher number of IOPs. However also note how the amount of CPU time used per IOP is lower on the right with the NVMe drive.

NVMe storage I/O performance
Figure 5 6 Gbps SATA NVM flash SSD vs. NVMe flash SSD

While many people are aware or learning about the IOP and bandwidth improvements as well as the decrease in latency with NVMe, something that gets overlooked is how much less CPU is used. If a server is spending time in wait modes that can result in lost productivity, by finding and removing the barriers more work can be done on a given server, perhaps even delaying a server upgrade.

In figure 5 notice the lower amount of CPU used per work activity being done (e.g. I/O or IOP) which translates to more effective resource use of your server. What that means is either doing more work with what you have, or potentially delaying a CPU server upgrade, or, using those extra CPU cycles to power software defined storage management stacks including erasure coding or advanced parity RAID, replication and other functions.

Table 1 shows relative server I/O performance of some NVM flash SSD devices across various workloads. As with any performance, the comparison takes them, and the following with a grain of salt as your speed will vary.

8KB I/O Size

1MB I/O size

NAND flash SSD

100% Seq. Read

100% Seq. Write

100% Ran. Read

100% Ran. Write

100% Seq. Read

100% Seq. Write

100% Ran. Read

100% Ran. Write

NVMe

IOPs

41829.19

33349.36

112353.6

28520.82

1437.26

889.36

1336.94

496.74

PCIe

Bandwidth

326.79

260.54

877.76

222.82

1437.26

889.36

1336.94

496.74

AiC

Resp.

3.23

3.90

1.30

4.56

178.11

287.83

191.27

515.17

CPU / IOP

0.001571

0.002003

0.000689

0.002342

0.007793

0.011244

0.009798

0.015098

12Gb

IOPs

34792.91

34863.42

29373.5

27069.56

427.19

439.42

416.68

385.9

SAS

Bandwidth

271.82

272.37

229.48

211.48

427.19

429.42

416.68

385.9

Resp.

3.76

3.77

4.56

5.71

599.26

582.66

614.22

663.21

CPU / IOP

0.001857

0.00189

0.002267

0.00229

0.011236

0.011834

0.01416

0.015548

6Gb

IOPs

33861.29

9228.49

28677.12

6974.32

363.25

65.58

356.06

55.86

SATA

Bandwidth

264.54

72.1

224.04

54.49

363.25

65.58

356.06

55.86

Resp.

4.05

26.34

4.67

35.65

704.70

3838.59

718.81

4535.63

CPU / IOP

0.001899

0.002546

0.002298

0.003269

0.012113

0.032022

0.015166

0.046545

Table 1 Relative performance of various protocols and interfaces

The workload results in table 1 were generated using a vdbench script running on a Windows 2012 R2 based server and are intended to be a relative indicator of different protocol and interfaces; your performance mileage will vary. The results shown below compare the number of IOPs (activity rate) for reads, writes, random and sequential across small 8KB and large 1MB sized I/Os.

Also shown in table 1 are bandwidth or throughput (e.g. amount of data moved), response time and the amount of CPU used per IOP. Note in table 1 how NVMe can do higher IOPs with a lower CPU per IOP, or, using a similar amount of CPU, do more work at a lower latency. SSD has been used for decades to help reduce CPU bottlenecks or defer server upgrades by removing I/O wait times and reduce CPU consumption (e.g. wait or lost time).

Can NVMe solutions run faster than those shown above? Absolutely!

Where To Learn More

View additional NVMe, SSD, NVM, SCM, Data Infrastructure and related topics via the following links.

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

Continue reading about NVMe with Part IV (Where and How to use NVMe) in this five-part series, or jump to Part I, Part II or Part V.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

Different NVMe Configurations

server storage I/O trends
Updated 1/12/2018

This is the second in a five-part mini-series providing a primer and overview of NVMe. View companion posts and more material at www.thenvmeplace.com.

The many different faces or facets of NVMe configurations

NVMe can be deployed and used in many ways, the following are some examples to show you its flexibility today as well as where it may be headed in the future. An initial deployment scenario is NVMe devices (e.g. PCIe cards, M2 or 8639 drives) installed as storage in servers or as back-end storage in storage systems. Figure 2 below shows a networked storage system or appliance that uses traditional server storage I/O interfaces and protocols for front-end access, however with back-end storage being all NVMe, or a hybrid of NVMe, SAS and SATA devices.
NVMe as back-end server storage I/O interface to NVM
Figure 2 NVMe as back-end server storage I/O interface to NVM storage

A variation of the above is using NVMe for shared direct attached storage (DAS) such as the EMC DSSD D5. In the following scenario (figure 3), multiple servers in a rack or cabinet configuration have an extended PCIe connection that attached to a shared storage all flash array using NVMe on the front-end. Read more about this approach and EMC DSSD D5 here or click on the image below.

EMC DSSD D5 NVMe
Figure 3 Shared DAS All Flash NVM Storage using NVMe (e.g. EMC DSSD D5)

Next up in figure 4 is a variation of the previous example, except NVMe is implemented over an RDMA (Remote Direct Memory Access) based fabric network using Converged 10GbE/40GbE or InfiniBand in what is known as RoCE (RDMA over Converged Ethernet pronounced Rocky).

NVMe over Fabric RoCE
Figure 4 NVMe as a “front-end” interface for servers or storage systems/appliances

Where To Learn More

View additional NVMe, SSD, NVM, SCM, Data Infrastructure and related topics via the following links.

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

Watch for more topology and configuration options as NVMe along with associated hardware, software and I/O networking tools and technologies emerge over time.

Continue reading about NVMe with Part III (Need for Performance Speed) in this five-part series, or jump to Part I, Part IV or Part V.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

3D XPoint nvm pm scm storage class memory

Part III – 3D XPoint server storage class memory SCM


Storage I/O trends

Updated 1/31/2018

3D XPoint nvm pm scm storage class memory.

This is the third of a three-part series on the recent Intel and Micron 3D XPoint server storage memory announcement. Read Part I here and Part II here.

What is 3D XPoint and how does it work?

3D XPoint is a new class or class of memory (view other categories of memory here) that provides performance for reads and writes closer to that of DRAM with about 10x the capacity density. In addition to the speed closer to DRAM vs. the lower NAND flash, 3D XPoint is also non-volatile memory (NVM) like NAND flash, NVRAM and others. What this means is that 3D XPoint can be used as persistent higher density fast server memory (or main memory for other computers and electronics). Besides being fast persistent main memory, 3D XPoint will also be a faster medium for solid state devices (SSD’s) including PCIe Add In Cards (AIC), m2 cards and drive form factor 8637/8639 NVM Express (NVMe) accessed devices that also has better endurance or life span compared to NAND flash.


3D XPoint architecture and attributes

The initial die or basic chip building block 3D XPoint implementation is a layer 128 Gbit device which if using 8 bits would yield 16GB raw. Over time increased densities should become available as the bit density improves with more cells and further scaling of the technology, combined with packaging. For example while a current die could hold up to 16 GBytes of data, multiple dies could be packaged together to create a 32GB, 64GB, 128GB etc. or larger actual product. Think about not only where packaged flash based SSD capacities are today, also think in terms of where DDR3 and DDR4 DIMM are at such as 4GB, 8GB, 16GB, 32GB densities.

The 3D aspect comes from the memory being in a matrix initially being two layers high, with multiple rows and columns that intersect, where those intersections occur is a microscopic material based switch for accessing a particular memory cell. Unlike NAND flash where an individual cell or bit is accessed as part of a larger block or page comprising several thousand bytes at once, 3D XPoint cells or bits can be individually accessed to speed up reads and writes in a more granular fashion. It is this more granular access along with performance that will enable 3D XPoint to be used in lower latency scenarios where DRAM would normally be used.

Instead of trapping electrons in a cell to create a bit of capacity (e.g. on or off) like NAND flash, 3D XPoint leverages the underlying physical material propertied to store a bit as a phase change enabling use of all cells. In other words, instead of being electron based, it is material based. While Intel and Micron did not specify what the actual chemistry and physical materials that are used in 3D XPoint, they did discuss some of the characteristics. If you want to go deep, check out how the Dailytech makes an interesting educated speculation or thesis on the underlying technology.

Watch the following video to get a better idea and visually see how 3D XPoint works.



3D XPoint YouTube Video

What are these chips, cells, wafers and dies?

Left many dies on a wafer, right, a closer look at the dies cut from the wafer

Dies (here and here) are the basic building block of what goes into the chips that in turn are the components used for creating DDR DIMM for main computer memory, as well as for create SD and MicroSD cards, USB thumb drives, PCIe AIC and drive form factor SSD, as well as custom modules on motherboards, or consumption via bare die and wafer level consumption (e.g. where you are doing really custom things at volume, beyond using a soldering iron scale).

Storage I/O trends

Has Intel and Micron cornered the NVM and memory market?

We have heard proclamations, speculation and statements of the demise of DRAM, NAND flash and other volatile and NVM memories for years, if not decades now. Each year there is the usual this will be the year of “x” where “x” can include among others. Resistive RAM aka ReRAM or RRAM aka the memristor that HP earlier announced they were going to bring to market and then earlier this year canceling those plans while Crossbar continues to pursue RRAM. MRAM or Magnetorestive RAM, Phase Change Memory aka CRAM or PCM and PRAM, FRAM aka FeRAM or Ferroelectric RAM among others.

flash SSD and NVM trends

Expanding persistent memory and SSD storage markets

Keep in mind that there are many steps taking time measured in years or decades to go from research and development lab idea to prototype that can then be produced at production volumes in economic yields. As a reference for, there is still plenty of life in both DRAM as well as NAND flash, the later having appeared around 1989.

Industry vs. Customer Adoption and deployment timeline

Technology industry adoption precedes customer adoption and deployment

There is a difference between industry adoption and deployment vs. customer adoption and deployment, they are related, yet separated by time as shown in the above figure. What this means is that there can be several years from the time a new technology is initially introduced and when it becomes generally available. Keep in mind that NAND flash has yet to reach its full market potential despite having made significant inroads the past few years since it was introduced in 1989.

This begs the question of if 3D XPoint is a variation of phase change, RRAM, MRAM or something else. Over at the Dailytech they lay out a line of thinking (or educated speculation) that 3D XPoint is some derivative or variation of phase change, time will tell about what it really is.

What’s the difference between 3D NAND flash and 3D XPoint?

3D NAND is a form of NAND flash NVM, while 3D XPoint is a completely new and different type of NVM (e.g. its not NAND).

3D NAND is a variation of traditional flash with the difference between vertical stacking vs. horizontal to improve density, also known as vertical NAND or V-NAND. Vertical stacking is like building up to house more tenants or occupants in a dense environment or scaling up, vs scaling-out by using up more space where density is not an issue. Note that magnetic HDD’s shifted to perpendicular (e.g. vertical) recording about ten years ago to break through the super parametric barrier and more recently, magnetic tape has also adopted perpendicular recording. Also keep in mind that 3D XPoint and the earlier announced Intel and Micron 3D NAND flash are two separate classes of memory that both just happen to have 3D in their marketing names.

Where to read, watch and learn more

Storage I/O trends

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

First, keep in mind that this is very early in the 3D XPoint technology evolution life-cycle and both DRAM and NAND flash will not be dead at least near term. Keep in mind that NAND flash appeared back in 1989 and only over the past several years has finally hit its mainstream adoption stride with plenty of market upside left. Same with DRAM which has been around for sometime, it too still has plenty of life left for many applications. However other applications that have the need for improved speed over NAND flash, or persistency and density vs. DRAM will be some of the first to leverage new NVM technologies such as 3D XPoint. Thus at least for the next several years, there will be a co-existences between new and old NVM and DRAM among other memory technologies. Bottom line, 3D XPoint is a new class of NVM memory, can be used for persistent main server memory or for persistent fast storage memory. If you have not done so, check out Part I here and Part II here of this three-part series on Intel and Micron 3D XPoint.

Disclosure: Micron and Intel have been direct and/or indirect clients in the past via third-parties and partners, also I have bought and use some of their technologies direct and/or in-direct via their partners.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

Intel Micron unveil new 3D XPoint Non Volatie Memory NVM for servers storage

3D XPoint NVM persistent memory PM storage class memory SCM


Storage I/O trends

Updated 1/31/2018

This is the first of a three-part series on Intel Micron unveil new 3D XPoint Non Volatie Memory NVM for servers storage announcement. Read Part II here and Part III here.

In a webcast the other day, Intel and Micron announced new 3D XPoint non-volatile memory (NVM) that can be used for both primary main memory (e.g. what’s in computers, serves, laptops, tablets and many other things) in place of Dynamic Random Access Memory (DRAM), for persistent storage faster than today’s NAND flash-based solid state devices (SSD), not to mention future hybrid usage scenarios. Note that this announcement while having the common term 3D in it is different from the earlier Intel and Micron announcement about 3D NAND flash (read more about that here).

Twitter hash tag #3DXpoint

The big picture, why this type of NVM technology is needed

Server and Storage I/O trends

  • Memory is storage and storage is persistent memory
  • No such thing as a data or information recession, more data being create, processed and stored
  • Increased demand is also driving density along with convergence across server storage I/O resources
  • Larger amounts of data needing to be processed faster (large amounts of little data and big fast data)
  • Fast applications need more and faster processors, memory along with I/O interfaces
  • The best server or storage I/O is the one you do not need to do
  • The second best I/O is one with least impact or overhead
  • Data needs to be close to processing, processing needs to be close to the data (locality of reference)


Server Storage I/O memory hardware and software hierarchy along with technology tiers

What did Intel and Micron announce?

Intel SVP and General Manager Non-Volatile Memory solutions group Robert Crooke (Left) and Micron CEO D. Mark Durcan did the joint announcement presentation of 3D XPoint (webinar here). What was announced is the 3D XPoint technology jointly developed and manufactured by Intel and Micron which is a new form or category of NVM that can be used for both primary memory in servers, laptops, other computers among other uses, as well as for persistent data storage.


Robert Crooke (Left) and Mark Durcan (Right)

Summary of 3D XPoint announcement

  • New category of NVM memory for servers and storage
  • Joint development and manufacturing by Intel and Micron in Utah
  • Non volatile so can be used for storage or persistent server main memory
  • Allows NVM to scale with data, storage and processors performance
  • Leverages capabilities of both Intel and Micron who have collaborated in the past
  • Performance Intel and Micron claim up to 1000x faster vs. NAND flash
  • Availability persistent NVM compared to DRAM with better durability (life span) vs. NAND flash
  • Capacity densities about 10x better vs. traditional DRAM
  • Economics cost per bit between dram and nand (depending on packaging of resulting products)

What applications and products is 3D XPoint suited for?

In general, 3D XPoint should be able to be used for many of the same applications and associated products that current DRAM and NAND flash-based storage memories are used for. These range from IT and cloud or managed service provider data centers based applications and services, as well as consumer focused among many others.


3D XPoint enabling various applications

In general, applications or usage scenarios along with supporting products that can benefit from 3D XPoint include among others’. Applications that need larger amounts of main memory in a denser footprint such as in-memory databases, little and big data analytics, gaming, wave form analysis for security, copyright or other detection analysis, life sciences, high performance compute and high-productivity compute, energy, video and content severing among many others.

In addition, applications that need persistent main memory for resiliency, or to cut delays and impacts for planned or un-planned maintenance or having to wait for memories and caches to be warmed or re-populated after a server boot (or re-boot). 3D XPoint will also be useful for those applications that need faster read and write performance compared to current generations NAND flash for data storage. This means both existing and emerging applications as well as some that do not yet exist will benefit from 3D XPoint over time, like how today’s applications and others have benefited from DRAM used in Dual Inline Memory Module (DIMM) and NAND flash advances over the past several decades.

Where to read, watch and learn more

Storage I/O trends

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

First, keep in mind that this is very early in the 3D XPoint technology evolution life-cycle and both DRAM and NAND flash will not be dead at least near term. Keep in mind that NAND flash appeared back in 1989 and only over the past several years has finally hit its mainstream adoption stride with plenty of market upside left. Continue reading Part II here and Part III here of this three-part series on Intel and Micron 3D XPoint along with more analysis and commentary.

Disclosure: Micron and Intel have been direct and/or indirect clients in the past via third-parties and partners, also I have bought and use some of their technologies direct and/or in-direct via their partners.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

I/O, I/O how well do you know good bad ugly server storage I/O iops?

How well do you know good bad ugly I/O iops?

server storage i/o iops activity data infrastructure trends

Updated 2/10/2018

There are many different types of server storage I/O iops associated with various environments, applications and workloads. Some I/Os activity are iops, others are transactions per second (TPS), files or messages per time (hour, minute, second), gets, puts or other operations. The best IO is one you do not have to do.

What about all the cloud, virtual, software defined and legacy based application that still need to do I/O?

If no IO operation is the best IO, then the second best IO is the one that can be done as close to the application and processor as possible with the best locality of reference.

Also keep in mind that aggregation (e.g. consolidation) can cause aggravation (server storage I/O performance bottlenecks).

aggregation causes aggravation
Example of aggregation (consolidation) causing aggravation (server storage i/o blender bottlenecks)

And the third best?

It’s the one that can be done in less time or at least cost or effect to the requesting application, which means moving further down the memory and storage stack.

solving server storage i/o blender and other bottlenecks
Leveraging flash SSD and cache technologies to find and fix server storage I/O bottlenecks

On the other hand, any IOP regardless of if for block, file or object storage that involves some context is better than those without, particular involving metrics that matter (here, here and here [webinar] )

Server Storage I/O optimization and effectiveness

The problem with IO’s is that they are a basic operations to get data into and out of a computer or processor, so there’s no way to avoid all of them, unless you have a very large budget. Even if you have a large budget that can afford an all flash SSD solution, you may still meet bottlenecks or other barriers.

IO’s require CPU or processor time and memory to set up and then process the results as well as IO and networking resources to move data too their destination or retrieve them from where they are stored. While IO’s cannot be eliminated, their impact can be greatly improved or optimized by, among other techniques, doing fewer of them via caching and by grouping reads or writes (pre-fetch, write-behind).

server storage I/O STI and SUT

Think of it this way: Instead of going on multiple errands, sometimes you can group multiple destinations together making for a shorter, more efficient trip. However, that optimization may also mean your drive will take longer. So, sometimes it makes sense to go on a couple of quick, short, low-latency trips instead of one larger one that takes half a day even as it accomplishes many tasks. Of course, how far you have to go on those trips (i.e., their locality) makes a difference about how many you can do in a given amount of time.

Locality of reference (or proximity)

What is locality of reference?

This refers to how close (i.e., its place) data exists to where it is needed (being referenced) for use. For example, the best locality of reference in a computer would be registers in the processor core, ready to be acted on immediately. This would be followed by levels 1, 2, and 3 (L1, L2, and L3) onboard caches, followed by main memory, or DRAM. After that comes solid-state memory typically NAND flash either on PCIe cards or accessible on a direct attached storage (DAS), SAN, or NAS device. 

server storage I/O locality of reference

Even though a PCIe NAND flash card is close to the processor, there still remains the overhead of traversing the PCIe bus and associated drivers. To help offset that impact, PCIe cards use DRAM as cache or buffers for data along with meta or control information to further optimize and improve locality of reference. In other words, this information is used to help with cache hits, cache use, and cache effectiveness vs. simply boosting cache use.

SSD to the rescue?

What can you do the cut the impact of IO’s?

There are many steps one can take, starting with establishing baseline performance and availability metrics.

The metrics that matter include IOP’s, latency, bandwidth, and availability. Then, leverage metrics to gain insight into your application’s performance.

Understand that IO’s are a fact of applications doing work (storing, retrieving, managing data) no matter whether systems are virtual, physical, or running up in the cloud. But it’s important to understand just what a bad IO is, along with its impact on performance. Try to identify those that are bad, and then find and fix the problem, either with software, application, or database changes. Perhaps you need to throw more software caching tools, hypervisors, or hardware at the problem. Hardware may include faster processors with more DRAM and faster internal busses.

Leveraging local PCIe flash SSD cards for caching or as targets is another option.

You may want to use storage systems or appliances that rely on intelligent caching and storage optimization capabilities to help with performance, availability, and capacity.

Where to gain insight into your server storage I/O environment

There are many tools that you can be used to gain insight into your server storage I/O environment across cloud, virtual, software defined and legacy as well as from different layers (e.g. applications, database, file systems, operating systems, hypervisors, server, storage, I/O networking). Many applications along with databases have either built-in or optional tools from their provider, third-party, or via other sources that can give information about work activity being done. Likewise there are tools to dig down deeper into the various data information infrastructure to see what is happening at the various layers as shown in the following figures.

application storage I/O performance
Gaining application and operating system level performance insight via different tools

windows and linux storage I/O performance
Insight and awareness via operating system tools on Windows and Linux

In the above example, Spotlight on Windows (SoW) which you can download for free from Dell here along with Ubuntu utilities are shown, You could also use other tools to look at server storage I/O performance including Windows Perfmon among others.

vmware server storage I/O
Hypervisor performance using VMware ESXi / vsphere built-in tools

vmware server storage I/O performance
Using Visual ESXtop to dig deeper into virtual server storage I/O performance

vmware server storage i/o cache
Gaining insight into virtual server storage I/O cache performance

Wrap up and summary

There are many approaches to address (e.g. find and fix) vs. simply move or mask data center and server storage I/O bottlenecks. Having insight and awareness into how your environment along with applications is important to know to focus resources. Also keep in mind that a bit of flash SSD or DRAM cache in the applicable place can go along way while a lot of cache will also cost you cash. Even if you cant eliminate I/Os, look for ways to decrease their impact on your applications and systems.

Where To Learn More

View additional NAS, NVMe, SSD, NVM, SCM, Data Infrastructure and HDD related topics via the following links.

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

>Keep in mind: SSD including flash and DRAM among others are in your future, the question is where, when, with what, how much and whose technology or packaging.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

Server and Storage IO Memory: DRAM and nand flash

Storage I/O trends

DRAM, DIMM, DDR3, nand flash memory, SSD, stating what’s often assumed

Often what’s assumed is not always the case. For example in along with around server, storage and IO networking circles including virtual as well as cloud environments terms such as nand (Negated AND or NOT And) flash memory aka (Solid State Device or SSD), DRAM (Dynamic Random Access Memory), DDR3 (Double Data Rate 3) not to mention DIMM (Dual Inline Memory Module) get tossed around with the assumption everybody must know what they mean.

On the other hand, I find plenty of people who are not sure what those among other terms or things are, sometimes they are even embarrassed to ask, particular if they are a self-proclaimed expert.

So for those who need a refresh or primer, here you go, an excerpt from Chapter 7 (Servers – Physical, Virtual and Software) from my book "The Green and Virtual Data Center" (CRC Press) available at Amazon.com and other global venues in print and ebook formats.

7.2.2 Memory

Computers rely on some form of memory ranging from internal registers, local on-board processor Level 1 (L1) and Level 2 (L2) caches, random accessible memory (RAM), non-volatile RAM (NVRAM) or nand Flash (SSD) along with external disk storage. Memory, which includes external disk storage, is used for storing operating system software along with associated tools or utilities, application programs and data. Main memory or RAM, also known as dynamic RAM (DRAM) chips, is packaged in different ways with a common form being dual inline memory modules (DIMMs) for notebook or laptop, desktop PC and servers.

RAM main memory on a server is the fastest form of memory, second only to internal processor or chip based registers, L1, L2 or local memory. RAM and processor based memories are volatile and non-persistent in that when power is removed, the contents of memory are lost. As a result, some form of persistent memory is needed to keep programs and data when power is removed. Read only memory (ROM) and NVRAM are both persistent forms of memory in that their contents are not lost when power is removed. The amount of RAM that can be installed into a server will vary with specific architecture implementation and operating software being used. In addition to memory capacity and packaging format, the speed of memory is also important to be able to move data and programs quickly to avoid internal bottlenecks. Memory bandwidth performance increases with the width of the memory bus in bits and frequency in MHz. For example, moving 8 bytes on a 64 bit buss in parallel at the same time at 100MHz provides a theoretical 800MByte/sec speed.

To improve availability and increase the level of persistence, some servers include battery backed up RAM or cache to protect data in the event of a power loss. Another technique to protect memory data on some servers is memory mirroring where twice the amount of memory is installed and divided into two groups. Each group of memory has a copy of data being stored so that in the event of a memory failure beyond those correctable with standard parity and error correction code (ECC) no data is lost. In addition to being fast, RAM based memories are also more expensive and used in smaller quantities compared to external persistent memories such as magnetic hard disk drives, magnetic tape or optical based memory medias.

Memory diagram
Memory and Storage Pyramid

The above shows a tiered memory model that may look familiar as the bottom part is often expanded to show tiered storage. At the top of the memory pyramid is high-speed processor memory followed by RAM, ROM, NVRAM and FLASH along with many forms of external memory commonly called storage. More detail about tiered storage is covered in chapter 8 (Data Storage – Data Storage – Disk, Tape, Optical, and Memory). In addition to being slower and lower cost than RAM based memories, disk storage along with NVRAM and FLASH based memory devices are also persistent.

By being persistent, when power is removed, data is retained on the storage or memory device. Also shown in the above figure is that on a relative basis, less energy is used for power storage or memory at the bottom of the pyramid than for upper levels where performance increases. From a PCFE (Power, Cooling, Floor space, Economic) perspective, balancing memory and storage performance, availability, capacity and energy to a given function, quality of service and service level objective for a given cost needs to be kept in perspective and not considering simply the lowest cost for the most amount of memory or storage. In addition to gauging memory on capacity, other metrics include percent used, operating system page faults and page read/write operations along with memory swap activity as well memory errors.

Base 2 versus base 10 numbering systems can account for some storage capacity that appears to “missing” when real storage is compared to what is expected to be seen. Disk drive manufacturers use base 10 (decimal) to count bytes of data while memory chip, server and operating system vendors typically use base 2 (binary) to count bytes of data. This has led to confusion when comparing a disk drive base 10 GB with a chip memory base 2 GB of memory capacity, such as 1,000,000,000 (10^9) bytes versus 1,073,741,824 (2^30) bytes. Nomenclature based on the International System of Units uses MiB, GiB and TiB to denote million, billion and trillion bytes for base 2 numbering with base 10 using MB, TB and GB . Most vendors do document how many bytes, sometimes in both base 2 and base 10, as well as the number of 512 byte sectors supported on their storage devices and storage systems, though it might be in the small print.

Related more reading:
How much storage performance do you want vs. need?
Can RAID extend the life of nand flash SSD?
Can we get a side of context with them IOPS and other storage metrics?
SSD & Real Estate: Location, Location, Location
What is the best kind of IO? The one you do not have to do
SSD, flash and DRAM, DejaVu or something new?

Ok, nuff said (for now).

Cheers
Gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press) and Resilient Storage Networks (Elsevier).

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO LLC All Rights Reserved

Posted on

Can RAID extend the life of nand flash SSD?

Storage I/O trends

Can RAID extend nand flash SSD life?

Imho, the short answer is YES, under some circumstances.

There is a myth and some FUD that RAID (Redundant Array of Independent Disks) can shorten the life durability of nand flash SSD (Solid State Device) vs. HDD (Hard Disk Drives) due to extra IOP’s. The reality is that depending on how configured, RAID level, implementation and other factors, nand flash SSD can be extended as I discuss in this here video.

Video

Nand flash SSD cells and wear

First, there is a myth that nand flash SSD does not have moving parts like hard disk drives (HDD’s) thus do not wear out or break. That is just a myth in that nand flash by its nature wears out with write usage. This is due to how they store data in cells that have a rated number of program erase (P/E) cycles that vary by type of medium. For example, Single Level Cell (SLC) has a longer P/E life duration vs. Multi-Level Cells (MLC) and eMLC that stack multiple cells together.

There are a number of factors that contribute to nand flash wear, also known as duty cycle or durability tied to P/E. For example, some storage systems or controllers do a better job both at the lower level flash translation layer (FTL) in addition to controllers, firmware, caching using DRAM and IO optimization such as write ordering or grouping.

Now what about this RAID and SSD thing?

Ok first as a recap keep in mind that there are many RAID levels along with variations, enhancements and where, or how implemented ranging from software to hardware, adapters to controllers to storage systems.

In the case of RAID 1 or mirroring, just like replication or other one to one or one too many copy operation a write to one device is echoed to another. In the case of RAID 5, data is spread across drives and parity; however, the parity is rotated across all drives in an equal manner.

Some FUD or myths or misunderstandings come into play is that not all RAID 5 implementations as an example are not the same. Some do a better job of buffering or caching data in battery protected mirrored DRAM memory until a full stripe write can occur, or if needed, a partial write.

Another attribute is the chunk or shard size (how much data is sent to each drive member) along with the stripe width (how many drives). Some systems have narrow stripes of say 3+1 or 4+1 or 5+1 while others can be 14+1 or 15+1 or wider. Thus, data can be written across a wider number of drives reducing the P/E consumption or use of a single drive depending on implementation.

How about RAID 6 (dual parity)?

Same thing, it is a matter of how well the implementation is, how the write gathering is done and so forth.

What about RAID wearing out nand flash SSD?

While it is possible that it has or can occur depending on type of RAID implementation, lack of caching or optimization, configuration, type of SSD, RAID level and other things, in general I will say myth busted.

Want some proof?

I could go through a long technical proof point and citing lots of facts, figures, experts and so forth leaving you all silenced and dazed similar to the students listening to Ben Stein in Ferris Buelers day off (Click here to see what I mean) asking “anybody anybody Buleler?

Ben Stein via https://nostagjicmoviesandthings.blogspot.com
Image via nostagjicmoviesandthings.blogspot.com

How about some simple SSD and storage math?

On a very conservative basis, my estimate is that around 250PB of nand flash SSD drives are shipped and installed on a revenue basis attached to or in storage systems and appliances. Combine what Dell + DotHill + EMC + Fujitsu + HDS + HP + IBM (including TMS) + NEC + NetApp + NEC + Oracle among other legacy along with new all flash as well as hybrid vendors (e.g. Cloudbyte, FusionIO (Via their Nexgen acquisition), Kaminario, Greenbytes, Nutanix or Nimble, Purestorage, Starboard or Solidfire, Tegile or Tintri, Violin or Whiptail among others).

It is also a safe assumption based on how customers configure and use those and other storage systems is with some form of RAID. Thus if things were as bad as some researchers were, vendors and their pundits have made them out to be, wouldn’t’t we be hearing of those issues?

Is it just a RAID 5 problem and that RAID 6 magically corrects the problem?

Well, that depends on apples to apples vs. apples to oranges comparisons.

For example if you are using a 14+2 (16 drive) RAID 6 to compare to say a 3+1 (4 drive) RAID 5 that is not a fair comparison. Granted, it is a handy one if you are a vendor that supports wider RAID groups, stripes and ranks vs. those who do not. However also keep in mind that some legacy vendors actually also support wide stripes and RAID groups.

So in some cases the magic is not in the RAID level, rather the implementation or how configured or lack thereof.

Video

Watch this TechTarget produced video recorded live while I was at EMCworld 2013 to learn more.

Otherwise, ok, nuff said (for now).

Cheers
Gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press) and Resilient Storage Networks (Elsevier)
twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO LLC All Rights Reserved

Posted on

Seagate provides proof of life: Enterprise HDD enhancements

Storage I/O trends

Proof of life: Enterprise Hard Disk Drives (HDD’s) are enhanced

Last week while hard disk drive (HDD) competitor Western Digital (WD) was announcing yet another (Velobit) in a string of acquisitions ( e.g. earlier included Stec, Arkeia) and investments (Skyera), Seagate announced new enterprise class HDD’s to their portfolio. Note that it was only two years ago that WD acquired Hitachi Global Storage Technologies (HGST) the disk drive manufacturing business of Hitachi Ltd. (not to be confused with HDS).

Seagate

Similar to WD expanding their presence in the growing nand flash SSD market, Seagate also in May of this year extended their existing enterprise class SSD portfolio. These enhancements included new drives with 12Gbs SAS interface, along with a partnership (and investment) with PCIe flash card startup vendor Virident. Other PCIe flash SSD card vendors (manufacturers and OEMs) include Cisco, Dell, EMC, FusionIO, HP, IBM, LSI, Micron, NetApp and Oracle among others.

These new Seagate enterprise class HDD’s are designed for use in cloud and traditional data center servers and storage systems. A month or two ago Seagate also announced new ultra-thin (5mm) client (aka desktop) class HDD’s along with a 3.5 inch 4TB video optimized HDD. The video optimized HDD’s are intended for Digital Video Recorders (DVR’s), Set Top Boxes (STB’s) or other similar applications.

What was announced?

Specifically what Seagate announced were two enterprise class drives, one for performance (e.g. 1.2TB 10K) and the other for space capacity (e.g. 4TB).

 

Enterprise High Performance 10K.7 (aka formerly known as Savio)

Enterprise Terascale (aka formerly known as constellation)

Class/category

Enterprise / High Performance

Enterprise High Capacity

Form factor

2.5” Small Form Factor (SFF)

3.5”

Interface

6Gbs SAS

6Gbs SATA

Space capacity

1,200GB (1.2TB)

4TB

RPM speed

10,000

5,900

Average seek

2.9 ms

12 ms

DRAM cache

64MB

64MB

Power idle / operating

4.8 watts

5.49 / 6.49 watts

Intelligent Power Management (IPM)

Yes – Seagate PowerChoice

Yes – Seagate PowerChoice

Warranty

Limited 5 years

Limited 3 years

Instant Secure Erase (ISE)

Yes

Optional

Other features

RAID Rebuild assist, Self-Encrypting Device (SED)

Advanced Format (AF) 4K block in addition to standard 512 byte sectors

Use cases

Replace earlier generation 3.5” 15K SAS and Fibre Channel HDD’s for higher performance applications including file systems, databases where SSD are not practical fit.

Backup and data protection, replication, copy operations for erasure coding and data dispersal, active in dormant archives, unstructured NAS, big data, data warehouse, cloud and object storage.

Note the Seagate Terascale has a disk rotation speed of 5,900 (5.9K RPM) which is not a typo given the more traditional 5.4K RPM drives. This slight increase in performance from 5.4K to 5.9K should give when combined with other enhancements (e.g. firmware, electronics) to boost performance for higher capacity workloads.

Let us watch for some performance numbers to be published by Seagate or others. Note that I have not had a chance to try these new drives yet, however look forward to getting my hands on them (among others) sometime in the future for a test drive to add to the growing list found here (hey Seagate and WD, that’s a hint ;) ).

What this all means?

Storage I/O trends

Wait, weren’t HDD’s supposed to be dead or dying?

Some people just like new and emerging things and thus will declare anything existing or that they have lost interest in (or their jobs need it) as old, boring or dead.

For example if you listen to some, they may say nand flash SSD are also dead or dying. For what it is worth, imho nand flash-based SSDs still have a bright future in front of them even with new technologies emerging as they will take time to mature (read more here or listen here).

However, the reality is that for at least the next decade, like them or not, HDD’s will continue to play a role that is also evolving. Thus, these and other improvements with HDD’s will be needed until current nand flash or emerging PCM (Phase Change Memory) among other forms of SSD are capable of picking up all the storage workloads in a cost-effective way.

Btw, yes, I am also a fan and user of nand flash-based SSD’s, in addition to HDD’s and see roles for both as being viable complementing each other for traditional, virtual and cloud environments.

In short, HDD’s will keep spinning (pun intended) for some time granted their roles and usage will also evolve similar to that of tape summit resources.

Storage I/O trends

With this announcement by Seagate along with other enhancements from WD show that the HDD will not only see its 60th birthday, (and here), it will probably also easily see its 70th and not from the comfort of a computer museum. The reason is that there is yet another wave of HDD improvements just around the corner including Shingled Magnetic Recording (SMR) (more info here) along with Heat Assisted Magnetic Recording (HAMR) among others. Watch for more on HAMR and SMR in future posts. With these and other enhancements, we should be able to see a return to the rapid density improvements with HDD’s observed during the mid to late 2000 era when Perpendicular recording became available.

What is up with this ISE stuff is that the same as what Xiotech (e.g. XIO) had?

Is this the same technology that Xiotech (now Xio) referred to the ISE the answer is no. This Seagate ISE is for fast secure erase of data on disk. The benefit of Instant Secure Erase (ISE) is to cut from hours or days the time required to erase a drive for secure disposal to seconds (or less). For those environments that already factor drives erase time as part of those overall costs, this can increase the useful time in service to help improve TCO and ROI.

Wait a minute, aren’t slower RPM’s supposed to be lower performance?

Some of you might be wondering or asking the question of wait, how can a 10,000 revolution per minute (10K RPM) HDD be considered fast vs. a 15K HDD, let alone SSD?

Storage I/O trends

There is a trend occurring with HDD’s that the old rules of IOPS or performance being tied directly to the size and rotational speed (RPM’s) of drives, along with their interfaces. This comes down to being careful to judge a book or in this case a drive by its cover. While RPM’s do have an impact on performance, new generation drives at 10K such as some 2.5” models are delivering performance equal to or better than earlier generation 3.5” 15K device’s.

Likewise, there are similar improvements with 5.4K devices vs. previous generation 7.2K models. As you will see in some of the results found here, not all the old rules of thumbs when it comes to drive performance are still valid. Likewise, keep those metrics that matter in the proper context.


Click on above image to see various performance results

For example as seen in the results (above), the more DRAM or DDR cache on the drives has a positive impact on sequential reads which can be good news if that is what your applications need. Thus, do your homework and avoid judging a device simply by its RPM, interface or form factor.

Other considerations, temperature and vibration

Another consideration is that with increased density of more drives being placed in a given amount of space, some of which may not have the best climate controls, humidity and vibration are concerns. Thus, the importance of drives having vibration dampening or safeguards to keep up performance are important. Likewise, even though drive heads and platters are sealed, there are also considerations that need to be taken care of for humidity in data center or cloud service providers in hot environments near the equator.

If this is not connecting with you, think about how close parts of Southeast Asia and the India subcontinent are to the equator along with the rapid growth and low-cost focus occurring there. Your data center might be temperature and humidity controlled, however others who very focused on cost cutting may not be as concerned with normal facilities best practices.

What type of drives should be used for cloud, virtual and traditional storage?

Good question and one where the answer should be it depends upon what you are trying or need to do (e.g. see previous posts here or here and here (via Seagate)).For example here are some tips for big data storage and storage making decisions in general.

Disclosure

Seagate recently invited me along with several other industry analysts to their cloud storage analyst summit in San Francisco where they covered roundtrip coach airfare, lodging, airport transfers and a nice dinner at the Epic Roast house.

hdd image

I also have received in the past a couple of Momentus XT HHDD (aka SSHD) from Seagate. These are in addition to those that I bought including various Seagate, WD along with HGST, Fujitsu, Toshiba and Samsung (SSD and HDD’s) that I use for various things.

Ok, nuff said (for now).

Cheers gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press) and Resilient Storage Networks (Elsevier)
twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO LLC All Rights Reserved

Posted on

Part II: How many IOPS can a HDD HHDD SSD do with VMware?

How many IOPS can a HDD HHDD SSD do with VMware?

server storage data infrastructure i/o iop hdd ssd trends

Updated 2/10/2018

This is the second post of a two-part series looking at storage performance, specifically in the context of drive or device (e.g. mediums) characteristics of How many IOPS can a HDD HHDD SSD do with VMware. In the first post the focus was around putting some context around drive or device performance with the second part looking at some workload characteristics (e.g. benchmarks).

A common question is how many IOPS (IO Operations Per Second) can a storage device or system do?

The answer is or should be it depends.

Here are some examples to give you some more insight.

For example, the following shows how IOPS vary by changing the percent of reads, writes, random and sequential for a 4K (4,096 bytes or 4 KBytes) IO size with each test step (4 minutes each).

IO Size for test
Workload Pattern of test
Avg. Resp (R+W) ms
Avg. IOP Sec (R+W)
Bandwidth KB Sec (R+W)
4KB
100% Seq 100% Read
0.0
29,736
118,944
4KB
60% Seq 100% Read
4.2
236
947
4KB
30% Seq 100% Read
7.1
140
563
4KB
0% Seq 100% Read
10.0
100
400
4KB
100% Seq 60% Read
3.4
293
1,174
4KB
60% Seq 60% Read
7.2
138
554
4KB
30% Seq 60% Read
9.1
109
439
4KB
0% Seq 60% Read
10.9
91
366
4KB
100% Seq 30% Read
5.9
168
675
4KB
60% Seq 30% Read
9.1
109
439
4KB
30% Seq 30% Read
10.7
93
373
4KB
0% Seq 30% Read
11.5
86
346
4KB
100% Seq 0% Read
8.4
118
474
4KB
60% Seq 0% Read
13.0
76
307
4KB
30% Seq 0% Read
11.6
86
344
4KB
0% Seq 0% Read
12.1
82
330

Dell/Western Digital (WD) 1TB 7200 RPM SATA HDD (Raw IO) thread count 1 4K IO size

In the above example the drive is a 1TB 7200 RPM 3.5 inch Dell (Western Digital) 3Gb SATA device doing raw (non file system) IO. Note the high IOP rate with 100 percent sequential reads and a small IO size which might be a result of locality of reference due to drive level cache or buffering.

Some drives have larger buffers than others from a couple to 16MB (or more) of DRAM that can be used for read ahead caching. Note that this level of cache is independent of a storage system, RAID adapter or controller or other forms and levels of buffering.

Does this mean you can expect or plan on getting those levels of performance?

I would not make that assumption, and thus this serves as an example of using metrics like these in the proper context.

Building off of the previous example, the following is using the same drive however with a 16K IO size.

IO Size for test
Workload Pattern of test
Avg. Resp (R+W) ms
Avg. IOP Sec (R+W)
Bandwidth KB Sec (R+W)
16KB
100% Seq 100% Read
0.1
7,658
122,537
16KB
60% Seq 100% Read
4.7
210
3,370
16KB
30% Seq 100% Read
7.7
130
2,080
16KB
0% Seq 100% Read
10.1
98
1,580
16KB
100% Seq 60% Read
3.5
282
4,522
16KB
60% Seq 60% Read
7.7
130
2,090
16KB
30% Seq 60% Read
9.3
107
1,715
16KB
0% Seq 60% Read
11.1
90
1,443
16KB
100% Seq 30% Read
6.0
165
2,644
16KB
60% Seq 30% Read
9.2
109
1,745
16KB
30% Seq 30% Read
11.0
90
1,450
16KB
0% Seq 30% Read
11.7
85
1,364
16KB
100% Seq 0% Read
8.5
117
1,874
16KB
60% Seq 0% Read
10.9
92
1,472
16KB
30% Seq 0% Read
11.8
84
1,353
16KB
0% Seq 0% Read
12.2
81
1,310

Dell/Western Digital (WD) 1TB 7200 RPM SATA HDD (Raw IO) thread count 1 16K IO size

The previous two examples are excerpts of a series of workload simulation tests (ok, you can call them benchmarks) that I have done to collect information, as well as try some different things out.

The following is an example of the summary for each test output that includes the IO size, workload pattern (reads, writes, random, sequential), duration for each workload step, totals for reads and writes, along with averages including IOP’s, bandwidth and latency or response time.

disk iops

Want to see more numbers, speeds and feeds, check out the following table which will be updated with extra results as they become available.

Device
Vendor
Make

Model

Form Factor
Capacity
Interface
RPM Speed
Raw
Test Result
HDD
HGST
Desktop
HK250-160
2.5
160GB
SATA
5.4K
HDD
Seagate
Mobile
ST2000LM003
2.5
2TB
SATA
5.4K
HDD
Fujitsu
Desktop
MHWZ160BH
2.5
160GB
SATA
7.2K
HDD
Seagate
Momentus
ST9160823AS
2.5
160GB
SATA
7.2K
HDD
Seagate
MomentusXT
ST95005620AS
2.5
500GB
SATA
7.2K(1)
HDD
Seagate
Barracuda
ST3500320AS
3.5
500GB
SATA
7.2K
HDD
WD/Dell
Enterprise
WD1003FBYX
3.5
1TB
SATA
7.2K
HDD
Seagate
Barracuda
ST3000DM01
3.5
3TB
SATA
7.2K
HDD
Seagate
Desktop
ST4000DM000
3.5
4TB
SATA
HDD
HDD
Seagate
Capacity
ST6000NM00
3.5
6TB
SATA
HDD
HDD
Seagate
Capacity
ST6000NM00
3.5
6TB
12GSAS
HDD
HDD
Seagate
Savio 10K.3
ST9300603SS
2.5
300GB
SAS
10K
HDD
Seagate
Cheetah
ST3146855SS
3.5
146GB
SAS
15K
HDD
Seagate
Savio 15K.2
ST9146852SS
2.5
146GB
SAS
15K
HDD
Seagate
Ent. 15K
ST600MP0003
2.5
600GB
SAS
15K
SSHD
Seagate
Ent. Turbo
ST600MX0004
2.5
600GB
SAS
SSHD
SSD
Samsung
840 PRo
MZ-7PD256
2.5
256GB
SATA
SSD
HDD
Seagate
600 SSD
ST480HM000
2.5
480GB
SATA
SSD
SSD
Seagate
1200 SSD
ST400FM0073
2.5
400GB
12GSAS
SSD

Performance characteristics 1 worker (thread count) for RAW IO (non-file system)

Note: (1) Seagate Momentus XT is a Hybrid Hard Disk Drive (HHDD) based on a 7.2K 2.5 HDD with SLC nand flash integrated for read buffer in addition to normal DRAM buffer. This model is a XT I (4GB SLC nand flash), may add an XT II (8GB SLC nand flash) at some future time.

As a starting point, these results are raw IO with file system based information to be added soon along with more devices. These results are for tests with one worker or thread count, other results will be added with such as 16 workers or thread counts to show how those differ.

The above results include all reads, all writes, mix of reads and writes, along with all random, sequential and mixed for each IO size. IO sizes include 4K, 8K, 16K, 32K, 64K, 128K, 256K, 512K, 1024K and 2048K. As with any workload simulation, benchmark or comparison test, take these results with a grain of salt as your mileage can and will vary. For example you will see some what I consider very high IO rates with sequential reads even without file system buffering. These results might be due to locality of reference of IO’s being resolved out of the drives DRAM cache (read ahead) which vary in size for different devices. Use the vendor model numbers in the table above to check the manufactures specs on drive DRAM and other attributes.

If you are used to seeing 4K or 8K and wonder why anybody would be interested in some of the larger sizes take a look at big fast data or cloud and object storage. For some of those applications 2048K may not seem all that big. Likewise if you are used to the larger sizes, there are still applications doing smaller sizes. Sorry for those who like 512 byte or smaller IO’s as they are not included. Note that for all of these unless indicated a 512 byte standard sector or drive format is used as opposed to emerging Advanced Format (AF) 4KB sector or block size. Watch for some more drive and device types to be added to the above, along with results for more workers or thread counts, along with file system and other scenarios.

Using VMware as part of a Server, Storage and IO (aka StorageIO) test platform

vmware vexpert

The above performance results were generated on Ubuntu 12.04 (since upgraded to 14.04 which was hosted on a VMware vSphere 5.1 (upgraded to 5.5U2) purchased version (you can get the ESXi free version here) with vCenter enabled system. I also have VMware workstation installed on some of my Windows-based laptops for doing preliminary testing of scripts and other activity prior to running them on the larger server-based VMware environment. Other VMware tools include vCenter Converter, vSphere Client and CLI. Note that other guest virtual machines (VMs) were idle during the tests (e.g. other guest VMs were quiet). You may experience different results if you ran Ubuntu native on a physical machine or with different adapters, processors and device configurations among many other variables (that was a disclaimer btw ;) ).

Storage I/O trends

All of the devices (HDD, HHDD, SSD’s including those not shown or published yet) were Raw Device Mapped (RDM) to the Ubuntu VM bypassing VMware file system.

Example of creating an RDM for local SAS or SATA direct attached device.

vmkfstools -z /vmfs/devices/disks/naa.600605b0005f125018e923064cc17e7c /vmfs/volumes/dat1/RDM_ST1500Z110S6M5.vmdk

The above uses the drives address (find by doing a ls -l /dev/disks via VMware shell command line) to then create a vmdk container stored in a dat. Note that the RDM being created does not actually store data in the .vmdk, it’s there for VMware management operations.

If you are not familiar with how to create a RDM of a local SAS or SATA device, check out this post to learn how.This is important to note in that while VMware was used as a platform to support the guest operating systems (e.g. Ubuntu or Windows), the real devices are not being mapped through or via VMware virtual drives.

vmware iops

The above shows examples of RDM SAS and SATA devices along with other VMware devices and dats. In the next figure is an example of a workload being run in the test environment.

vmware iops

One of the advantages of using VMware (or other hypervisor) with RDM’s is that I can quickly define via software commands where a device gets attached to different operating systems (e.g. the other aspect of software defined storage). This means that after a test run, I can quickly simply shutdown Ubuntu, remove the RDM device from that guests settings, move the device just tested to a Windows guest if needed and restart those VMs. All of that from where ever I happen to be working from without physically changing things or dealing with multi-boot or cabling issues.

Where To Learn More

View additional NAS, NVMe, SSD, NVM, SCM, Data Infrastructure and HDD related topics via the following links.

Additional learning experiences along with common questions (and answers), as well as tips can be found in Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What This All Means

So how many IOPs can a device do?

That depends, however have a look at the above information and results.

Check back from time to time here to see what is new or has been added including more drives, devices and other related themes.

Ok, nuff said, for now.

Gs

Greg Schulz – Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2017 (vSAN and vCloud). Author of Software Defined Data Infrastructure Essentials (CRC Press), as well as Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press), Resilient Storage Networks (Elsevier) and twitter @storageio. Courteous comments are welcome for consideration. First published on https://storageioblog.com any reproduction in whole, in part, with changes to content, without source attribution under title or without permission is forbidden.

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of Server StorageIO.

Posted on

FusionIO (FIO) SSD vendor CEO out in a flash, whats up with that?

Storage I/O trends

FusionIO (FIO) who recently bought Nexgen to expand their reach from just a server centric to a more broad flash focus has seen their CEO and founder David Flynn race out the door. Not surprisingly, wall street who does not like to be surprised were surprised just a week or two after the most recent earning announcements reacted with a sell off of the FIO stock.

Here is the conundrum, those who were or are fans of Flynn, FIO and their approach along with server centric in your face approach may not be happy with this move.

On the other hand, those were not fans of Flynn, FIO and their approach of getting in your face of having others do so if you did not fall into their ranks may be happy with this move.

One question is was Flynn shown the door and left before it could hit his backside on the way out, or, did he see something and pulled the rip cord on his golden parachute, or some other or combination?

With the recent Nexgen acquisition which could be seen as a move by FIO (and their board of directors) to make more attractive either for an acquisition. Or, to transition from a server-side centric approach to a broader focus.

If the former, perhaps Flynn sees or saw the writing on the wall on who those suitors might or would be and decided to take his money now and run joining the serial entrepreneur ranks now.

Otoh, perhaps Flynn was just too focused with a singular focus and passion on the server space thus not able or interested in transitioning to a broader focus, which might also have involved eating a bit of crow. By eating a bit of crow, I mean given some of the in your face and it’s the FIO way or the highway approach of server only flash.

With Nexgen to be successful that would involve aligning more with the larger vendors and other startups who offer broader portfolios, something that was targeted and mud or fud thrown at by FIO, something that some CEOs or others can have challenges with. It should also be noted that FIO has brought in new employees with experience in broader marketers, not to mention industry veterans like John Spiers of Nexgen.

Candidly, I am not sure which of the above is the scenario, however, for those involved with FIO as employees, partners, customers and shareholders I hope some clarity arrives soon for them. Whether that clarity is via an acquisition (who is one of many questions), or a launching as FIO 2.0 or something similar with a focus on bring more capabilities to customers, increasing their touch points selling more products, hardware, software as opposed to leaving those for others (e.g. their competitors).

Ok, nuff said (for now)

Cheers gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press), The Green and Virtual Data Center (CRC Press) and Resilient Storage Networks (Elsevier)
twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO LLC All Rights Reserved

Posted on

NetApp EF540, something familiar, something new

StorageIO Industry trends and perspectives image

NetApp announced the other day a new all nand flash solid-state devices (SSD) storage system called the EF540 that is available now. The EF540 has something’s new and cool, along with some things familiar, tried, true and proven.

What is new is that the EF540 is an all nand flash multi-level cell (MLC) SSD storage system. What is old is that the EF540 is based on the NetApp E-Series (read more here and here) and SANtricity software with hundreds of thousands installed systems. As a refresher, the E-Series are the storage system technologies and solutions obtained via the Engenio acquisition from LSI in 2011.

Image of NetApp EF540 via ntapgeek.com
Image via www.ntapgeek.com

The EF540 expands the NetApp SSD flash portfolio which includes products such as FlashCache (read cache aka PAM) for controllers in ONTAP based storage systems. Other NetApp items in the NetApp flash portfolio include FlashPool SSD drives for persistent read and write storage in ONTAP based systems. Complimenting FlashCache and FlashPool is the server-side PCIe caching card and software FlashAccel. NetApp is claiming to have revenue shipped 36PB of flash complimenting over 3 Exabytes (EB) of storage while continuing to ship a large amount of SAS and SATA HDD’s.

NetApp also previewed its future FlashRay storage system that should appear in beta later in 2013 and general availability in 2014.

In addition to SSD and flash related announcements, NetApp also announced enhancements to its ONTAP FAS/V6200 series including the FAS/V6220, FAS/V6250 and FAS/V6290.

Some characteristics of the NetApp EF540 and SANtricity include:

  • Two models with 12 or 24 x 6Gbs SAS 800GB MLC SSD devices
  • Up to 9.6TB or 19.2TB physical storage in a 2U (3.5 inch) tall enclosure
  • Dual controllers for redundancy, load-balancing and availability
  • IOP performance of over 300,000 4Kbyte random 100% reads under 1ms
  • 6GByte/sec performance of 512Kbyte sequential reads, 5.5Gbyte/sec random reads
  • Multiple RAID levels (0, 1, 10, 3, 5, 6) and flexible group sizes
  • 12GB of DRAM cache memory in each controller (mirrored)
  • 4 x 8GFC host server-side ports per controller
  • Optional expansion host ports (6Gb SAS, 8GFC, 10Gb iSCSI, 40Gb IBA/SRP)
  • Snapshots and replication (synchronous and asynchronous) including to HDD systems
  • Can be used for traditional IOP intensive little-data, or bandwidth for big-data
  • Proactive SSD wear monitoring and notification alerts
  • Utilizes SANtricity version 10.84

Poll, Are large storage arrays day’s numbered?

EMC and NetApp (along with other vendors) continue to sell large numbers of HDD’s as well as large amounts of SSD. Both EMC and NetApp are taking similar approaches of leveraging PCIe flash cards as cache adding software functionality to compliment underlying storage systems. The benefit is that the cache approach is less disruptive for many environments while allowing improved return on investment (ROI) of existing assets.

EMC

NetApp

Storage systems with HDD and SSD

VMAX, VNX

FAS/V, E-Series

Storage systems with SSD cache

FastCache,

FlashCache

All SSD based storage

VMAX, VNX

EF540

All new SSD system in development

Project X

FlashRay

Server side PCIe SSD cache

VFCache

FlashAcell

Partner ecosystems

Yes

Yes

The best IO is the one that you do not have to do, however the next best are those that have the least cost or affect which is where SSD comes into play. SSD is like real estate in that location matters in terms of providing benefit, as well as how much space or capacity is needed.

What does this all mean?
The NetApp EF540 based on the E-Series storage system architecture is like one of its primary competitors (e.g. EMC VNX also available as an all-flash model). The similarity is that both have been competitors, as well as have been around for over a decade with hundreds of thousands of installed systems. The similarities are also that both continue to evolve their code base leveraging new hardware and software functionality. These improvements have resulted in improved performance, availability, capacity, energy effectiveness and cost reduction.

Whats your take on RAID still being relevant?

From a performance perspective, there are plenty of public workloads and benchmarks including Microsoft ESRP and SPC among others to confirm its performance. Watch for NetApp to release EF540 SPC results given their history of doing so with other E-Series based systems. With those or other results, compare and contrast to other solutions looking not just at IOPS or MB/sec (bandwidth), also latency, functionality and cost.

What does the EF540 compete with?
The EF540 competes with all flash-based SSD solutions (Violin, Solidfire, Purestorage, Whiptail, Kaminario, IBM/TMS, up-coming EMC Project “X” (aka XtremeIO)) among others. Some of those systems use general-purpose servers combined SSD drives, PCIe cards along with management software where others leverage customized platforms with software. To a lesser extent, competition will also be mixed mode SSD and HDD solutions along with some PCIe target SSD cards for some situations.

What to watch and look for:
It will be interesting to view and contrast public price performance results using SPC or Microsoft ESRP among others to see how the EF540 compares. In addition, it will be interesting to compare other storage based, as well as SSD systems beyond the number of IOPS. What will be interesting is to keep an eye on latency, as well as bandwidth, feature functionality and associated costs.

Given that the NetApp E-Series are OEM or sold by third parties, let’s see if something looking similar or identical to the EF540 appear at any of those or new partners. This includes traditional general purpose and little-data environments, along with cloud, managed service provider, high performance compute and high productivity compute (HPC), super computer (SC), big data and big bandwidth among others.

Poll, Have SSD been successful in traditional storage systems and arrays

The EF540 could also appear as a storage or IO accelerator for large-scale out, clustered, grid and object storage systems for meta data, indices, key value stores among other uses either direct attached to servers, or via shared iSCSI, SAS, FC and InfiniBand (IBA) SCSI Remote Protocol (SRP).

Keep an eye on how the startups that have been primarily Just a Bunch Of SSD (JBOS) in a box start talking about adding new features and functionality such as snapshots, replication or price reductions. Also, keep an eye and ear open to what EMC does with project “X” along with NetApp FlashRay among other improvements.

For NetApp customers, prospects, partners, E-Series OEMs and their customers with the need for IO consolidation, or performance optimization for big-data, little-data and related applications the EF540 opens up new opportunities and should be good news. For EMC competitors, they now have new competition which also signals an expanding market with new opportunities in adjacent areas for growth. This also further signals the need for diverse ssd portfolios and product options to meet different customer application needs, along with increased functionality vs. lowest cost for high capacity fast nand SSD storage.

Some related reading:

Disclosure: NetApp, Engenio (when LSI), EMC and TMS (now IBM) have been clients of StorageIO.

Ok, nuff said

Cheers gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press, 2011), The Green and Virtual Data Center (CRC Press, 2009), and Resilient Storage Networks (Elsevier, 2004)

twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2024 Server StorageIO and UnlimitedIO LLC All Rights Reserved

Posted on

SSD past, present and future with Jim Handy

Now also available via

This is a new episode in the continuing StorageIO industry trends and perspectives pod cast series (you can view more episodes or shows along with other audio and video content here) as well as listening via iTunes or via your preferred means using this RSS feed (https://storageio.com/StorageIO_Podcast.xml)

StorageIO industry trends cloud, virtualization and big data

In this episode, I talk with SSD nand flash and DRAM chip analyst Jim Handy of Objective Analysis at the LSI AIS (Accelerating Innovation Summit) 2012 in San Jose. Our conversation includes SSD past, present and future, market and industry trends, who are doing what and things to keep an eye and ear, open for along with server, storage and memory convergence.

Click here (right-click to download MP3 file) or on the microphone image to listen to the conversation with Jim and myself.

StorageIO podcast

Also available via

Watch (and listen) for more StorageIO industry trends and perspectives audio blog posts pod casts and other upcoming events. Also be sure to heck out other related pod casts, videos, posts, tips and industry commentary at StorageIO.com and StorageIOblog.com.

Enjoy this episode SSD Past, Present and Future with Jim Handy.

Ok, nuff said.

Cheers gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press, 2011), The Green and Virtual Data Center (CRC Press, 2009), and Resilient Storage Networks (Elsevier, 2004)

twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2012 StorageIO and UnlimitedIO All Rights Reserved

Posted on

IBM buys flash solid state device (SSD) industry veteran TMS

How much flash (or DRAM) based Solid State Device (SSD) do you want or need?

IBM recently took a flash step announcing it wants and needs more SSD capabilities in different packaging and functionality capabilities to meet the demands and opportunities of customers, business partners and prospects by acquiring Texas Memory Systems (TMS).

IBM buys SSD flash vendor TMS

Unlike most of the current generation of SSD vendors besides those actually making the dies (chips or semiconductors) or SSD drives that are startups or relatively new, TMS is the industry veteran. Where most of the current SSD vendors experiences (as companies) is measured in months or at best years, TMS has seen several generations and SSD adoption cycles during its multi-decade existence.

IBM buys SSD vendor Texas Memory Systems TMS

What this means is that TMS has been around during past dynamic random access memory (DRAM) based SSD cycles or eras, as well as being an early adopter and player in the current nand flash SSD era or cycle.

Granted, some in the industry do not consider the previous DRAM based generation of products as being SSD, and vice versa, some DRAM era SSD aficionados do not consider nand flash as being real SSD. Needless to say that there are many faces or facets to SSD ranging in media (DRAM, and nand flash among others) along with packaging for different use cases and functionality.

IBM along with some other vendors recognize that the best type of IO is the one that you do not have to do. However reality is that some type of Input Output (IO) operations need to be done with computer systems. Hence the second best type of IO is the one that can be done with the least impact to applications in a cost-effective way to meet specific service level objectives (SLO) requirements. This includes leveraging main memory or DRAM as cache or buffers along with server-based PCIe SSD flash cards as cache or target devices, along with internal SSD drives, as well as external SSD drives and SSD drives and flash cards in traditional storage systems or appliances as well as purpose-built SSD storage systems.

While TMS does not build the real nand flash single level cell (SLC) or multi-level cell (MLC) SSD drives (like those built by Intel, Micron, Samsung, SANdisk, Seagate, STEC and Western Digital (WD) among others), TMS does incorporate nand flash chips or components that are also used by others who also make nand flash PCIe cards and storage systems.

StorageIO industry trend for storage IO

IMHO this is a good move for both TMS and IBM, both of whom have been StorageIO clients in the past (here, here and here) that was a disclosure btw ;) as it gives TMS, their partners and customers a clear path and large organization able to invest in the technologies and solutions on a go forward basis. In other words, TMS who had looked to be bought gets certainty about their future as do they clients.

IBM who has used SSD based components such as PCIe flash SSD cards and SSD based drives from various suppliers gets a PCIe SSD card of their own, along with purpose-built mature SSD storage systems that have lineages to both DRAM and nand flash-based experiences. Thus IBM controls some of their own SSD intellectual property (e.g. IP) for PCIe cards that can go in theory into their servers, as well as storage systems and appliances that use Intel based (e.g. xSeries from IBM) and IBM Power processor based servers as a platform such. For example DS8000 (Power processor), and Intel based XIV, SONAS, V7000, SVC, ProtecTier and Pursystems (some are Power based).

In addition IBM also gets a field proven purpose-built all SSD storage system to compete with those from startups (Kaminario, Purestorage, Solidfire, Violin and Whiptail among others), as well as those being announced from competitors such as EMC (e.g. project X and project thunder) in addition to SSD drives that can go into servers and storage systems.

The question should not be if SSD is in your future, rather where will you be using it, in the server or a storage system, as a cache or a target, as a PCIe target or cache card or as a drive or as a storage system. This also means the question of how much SSD do you need along with what type (flash or DRAM), for what applications and how configured among other topics.

Storage and Memory Hirearchy diagram where SSD fits

What this means is that there are many locations and places where SSD fits, one type of product or model does not fit or meet all requirements and thus IBM with their acquisition of TMS, along with presumed partnership with other SSD based components will be able to offer a diverse SSD portfolio.

StorageIO industry trend for storage IO

The industry trend is for vendors such as Cisco, Dell, EMC, IBM, HP, NetApp, Oracle and others all of whom are either physical server and storage vendors, or in the case of EMC, virtual servers partnered with Cisco (vBlock and VCE) and Lenovo for physical servers.

Different types and locations for SSD

Thus it only makes sense for those vendors to offer diverse SSD product and solution offerings to meet different customer and application needs vs. having a single solution that users adapt to. In other words, if all you have is a hammer, everything needs to look like a nail, however if you have a tool box of various technologies, then it comes down to being able to leverage including articulating what to use when, where, why and how for different situations.

I think this is a good move for both IBM and TMS. Now lets watch how IBM and TMS can go beyond the press release, slide decks and webex briefings covering why it is a good move to justify their acquisition and plans, moving forward and to see the results of what is actually accomplished near and long-term.

Read added industry trends and perspective commentary about IBM buying TMS here and here, as well as check out these related posts and content:

How much SSD do you need vs. want?
What is the best kind of IO? The one you do not have to do
Is SSD dead? No, however some vendors might be
Has SSD put Hard Disk Drives (HDDs) On Endangered Species List?
Why SSD based arrays and storage appliances can be a good idea (Part I)
EMC VFCache respinning SSD and intelligent caching (Part I)
SSD options for Virtual (and Physical) Environments: Part I Spinning up to speed on SSD
Speaking of speeding up business with SSD storage
Is SSD dead? No, however some vendors might be
Part I: PureSystems, something old, something new, something from big blue
The Many Faces of Solid State Devices/Disks (SSD)
SSD and Green IT moving beyond green washing

Meanwhile, congratulations to both IBM and TMS, ok, nuff said (for now).

Cheers Gs

Greg Schulz – Author Cloud and Virtual Data Storage Networking (CRC Press, 2011), The Green and Virtual Data Center (CRC Press, 2009), and Resilient Storage Networks (Elsevier, 2004)

twitter @storageio

All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2012 StorageIO and UnlimitedIO All Rights Reserved