Category: nvme

Posted on

Installing new Insider’s build Windows Server vnext NVMeoF, host initiator

Installing new Insider’s build Windows Server vnext NVMeoF, host initiator

A few weeks ago, Microsoft released a new Insider’s build of Windows Server vnext with NVMeoF host initiator. This post looks at installing and setting up the new initiator for testing. Granted, this is for testing and evaluating the NVMe over Fabric (NVMeoF) that supports TCP and RDMA for networked block storage with a Windows Server vnext host, and a Linux test target. Given that this is for a simple test and trying things out, you will want to adjust some of the security and access control settings on both the initiator and target sides.

Whats Needed

For the host NVMeoF initiator, you will need a current version of the insider build release of Windows Server vnext available here. You will also need some type of NVMeoF target, such as a storage system or platform that supports NVMeoF via TCP or RDMA, or a test initiator, for example, on a Linux system. For my test target, I set up an Ubuntu 25 Linux server with a small 50GB device to export as an NVMeoF device.

On the Ubuntu Linux server (e.g NVMeoF target)

sudo apt install nvme-cli -y
sudo modprobe nvmet
sudo modprobe nvmet-tcp
lsmod | grep nvmet

You should then see something like the following:

nvmet_tcp 45056 1
nvmet 249856 7 nvmet_tcp
nvme_core 241664 4 nvmet,nvme_tcp,nvme,nvme_fabrics
nvme_keyring 20480 5 nvmet,nvme_tcp,nvmet_tcp,nvme_core,nvme_fabrics
nvme_auth 28672 2 nvmet,nvme_core

Continue with the NVMeoF target setup:

sudo mount -t configfs none /sys/kernel/config
cd /sys/kernel/config/nvmet

# define your nqn which can be what ever you like, or have

sudo mkdir subsystems/nqn-2026-03.com:nvme.storageio01

# since this is for testing, this is being loose with access control

echo 1 | sudo tee subsystems/nqn-2026-03.com:nvme.storageio01/attr_allow_any_host
sudo mkdir subsystems/nqn-2026-03.com:nvme.storageio01/namespaces/1

# specify device to use as target, the device could be a partition, volume, file, etc

echo /dev/sdb | sudo tee subsystems/nqn-2026-03.com:nvme.storageio01/namespaces/1/device_path
echo 1 | sudo tee subsystems/nqn-2026-03.com:nvme.storageio01/namespaces/1/enable

# You can select what port you want to use. For testing, used 4420

sudo mkdir ports/4420
echo tcp | sudo tee ports/4420/addr_trtype
echo 192.168.1.205 | sudo tee ports/4420/addr_traddr
echo 4420 | sudo tee ports/4420/addr_trsvcid
echo ipv4 | sudo tee ports/4420/addr_adrfam

# For non-testing, you can adjust access to suit your needs

sudo ufw allow 4420/tcp

# now make the target accessible

sudo ln -s /sys/kernel/config/nvmet/subsystems/nqn-2026-03.com:nvme.storageio01 /sys/kernel/config/nvmet/ports/4420/subsystems/nqn-2026-03.com:nvme.storageio01

# now lets see if there is a listener

sudo ss -ltnp | grep 4420

# Should see something like this:

LISTEN 0 128 192.168.1.205:4420 0.0.0.0:*

# now lets discover whats out there based on what we have setup

sudo nvme discover -t tcp -a 192.168.1.205 -s 4420

# should see something like this:

Discovery Log Number of Records 2, Generation counter 2
=====Discovery Log Entry 0======
trtype: tcp
adrfam: ipv4
subtype: current discovery subsystem
treq: not specified, sq flow control disable supported
portid: 4420
trsvcid: 4420
subnqn: nqn.2014-08.org.nvmexpress.discovery
traddr: 192.168.1.205
eflags: none
sectype: none
=====Discovery Log Entry 1======
trtype: tcp
adrfam: ipv4
subtype: nvme subsystem
treq: not specified, sq flow control disable supported
portid: 4420
trsvcid: 4420
subnqn: nqn-2026-03.com:nvme.storageio01
traddr: 192.168.1.205
eflags: none
sectype: none

# if you want to connect from the Linux server

sudo nvme connect -t tcp -a 192.168.1.205 -s 4420 -n nqn-2026-03.com:nvme.storageio01

# and then list the targets

nvme list

# you should see something like this:

Node Generic SN Model Namespace Usage Format FW Rev
——————— ——————— ——————– —————————————- ———- ————————– —————- ——–
/dev/nvme0n1 /dev/ng0n1 eab7bd175e01b9dbbba5 Linux 0x1 53.69 GB / 53.69 GB 512 B + 0 B 6.17.0-1

setting up linux nvmeof target

On Windows Server vnext (NVMeoF Initiator)

Now, on your Windows Server vnext, open a command window with elevated access (e.g., Run as Administrator) and use the following, making note of the parameters from the target (e.g., IP address, port 4420, nqn).

Here is a screen shot of the host initiator before the NVMeoF target appears.

windows server vnext nvmeof host initiator setup

Now display available NVMeoF initiators using this command:

nvmeofutil.exe list -t ia

windows server vnext nvmeof host initiator setup

Note the host adapter number (e.g., 1) which will be used in the next step to get host gateway information.

nvmeofutil.exe list -t sp -ia 1
windows server vnext nvmeof host initiator setup display host gateway

Note the host gateway, target IP address, target port, and target nqn. Also note that there is an IPV6 and IPV4, in this example there is no IPV6 being used, so we are using the 2nd item (e.g. IPV4).

nvmeofutil.exe add -t sp -ia 1 -hg -dy true -pi 1 -nq nqn-2026-03.com:nvme.storageio01 -ta 192.168.1.205 -ts 4420

windows server vnext nvmeof host initiator setup showing host gateway and other information

Now, let’s identify the target subsystem ID.

nvmeofutil.exe list -t sp -ia 1

windows server vnext nvmeof host initiator setup showing subsystem information of available targets

Now, let’s use the subsystem port from above to connect to the target. Note that in some of the instructions I have seen, there is a reference to sp without an sp shown; thus, the previous steps help identify the sp to use.

nvmeofutil.exe connect -ia 1 -sp 000000003C7E88F0

windows server vnext nvmeof host initiator setup

Congratulations, at this point, you should see a new disk or volume appear on your Windows Server vnext host system.

windows server vnext nvmeof host initiator setup

The above image shows what appears when using diskpart, as well as the disk format utility, or use your preferred command, PowerShell script, or tool. If the disk or volume has not yet been used, it must be initialized and formatted. Otoh, as in this example, the disk volume had previously been initialized and formatted as a simple NTFS volume.

Again, reiterating, the above is for simple testing and evaluation, with loose security and persistence that you will want to tighten up as needed. The whole process is pretty easy as long as you don’t make any typos, for example, with nqn or other items.

Additional Resources Where to learn more

Microsoft Windows Server Insider Build NVMeoF Initiator (Blog post)
NVMe Linux Driver and related info
Announcing Native NVMe in Windows Server 2025 (Microsoft Post)
Introducing the Windows NVMe-oF Initiator Preview in Windows Server Insiders Builds (Microsoft Post)
Windows Server Insider Builds (Microsoft Downloads)
ToE NVMeoF TCP Performance Boost Performance Reduce Costs (blog post)
TheNVMeplace.com (Various NVMe resources)
Azure Cloud Storage here (Microsoft Post)
Microsoft Azure Data Box (Blog Post)
Azure Elastic SAN from Cloud to On-Prem here (Blog Post)
Cloud and Software Defined Data Infrastructure topics here

Additional learning experiences along with common questions (and answers) pertaining to SCSI, Fibre Channel, NVMe, Cloud, Hardware, Software, Services, techniques, server, storage, I/O networking, data protection among other topics are found in my Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials (CRC Press) by Greg Schulz

What this all means

With the insiders preview of Windows Server vnext you have the opportunity to try out the new NVMeoF initiator using TCP and RDMA including on Ethernet based fabric networks. The current insiders build utilizes the nvmeofutility.exe for initial testing, evaluations and experiments while future releases should see other management tools and interfaces such as Powershell cmdlets among others.

Ok, nuff said.

Cheers Gs

Greg Schulz – Ten time Microsoft MVP Cloud and Data Center Management and Azure Storage, along with previous ten-time VMware vExpert. 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 2006-2026 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of UnlimitedIO LLC.

Posted on

Its Time for NVMeoF to let iSCSI begin its slow retirement journey

Why Its Time for NVMeoF

Its Time for NVMeoF to let iSCSI begin its slow retirement journey for networked shared block storage. Dating back to the late 1990s and early 2000s, iSCSI encapsulates the SCSI command set over a TCP/IP (IP)- based fabric network, or as an alternative to SCSI on Fibre Channel (SCSI_FCP aka FCP aka FC). The SCSI command set dates back to the 1980s, having been deployed on physical parallel cabling and later on other transports, including Fibre Channel, Serial Attached SCSI (SAS), and iSCSI.

The reason it’s time for NVMe over Fabric (NVMeoF), and specifically NVMeoF using TCP and RDMA, to let iSCSI begin its slow retirement journey, is the evolution of the NVMe protocol command set and the expansion of its ecosystem as a modern, ground-up, streamlined server storage I/O protocol command set.

Keep in mind that there are two aspects in play with NVMeoF, which are similar to iSCSI and Fibre Channel-based FCP: the upper layer or protocol command set (e.g., NVMe and SCSI), and media through which these command sets are implemented via various interfaces such as TCP or RDMA on Ethernet, via InfiniBand, Fibre Channel, or physical PCIe, among others.

Ethernet has established itself as the dominant network interface, supporting various upper layers, such as TCP, and increasing its speed. Likewise, InfiniBand and Fibre Channel continue to increase their speeds and maintain their deployment roles in some environments.

Besides supporting faster networks, the NVMe command set was architected and has been deployed to support the current and evolving generations of processors with multiple cores, enabling more concurrent I/O operations and compute threads with less CPU overhead due to a modern, streamlined I/O stack and associated native drivers.

Why NVMeoF (over TCP or RDMA)

Leverage the combined improved benefits of NVMe as a server storage I/O communications protocol, along with TCP networking and industry-standard Ethernet as a communications fabric. With the NVMe command protocol server storage, I/O processing is streamlined and optimized for modern high-speed, low-latency network fabrics, such as Ethernet, as the transport.

The NVMe command set, compared to the SCSI command set, enables faster I/O processing with lower latency, higher IOPs, greater bandwidth, and more transactions, while increasing concurrency and the number of threads and reducing CPU usage per I/O operation. Similar to the SCSI command set, which has been deployed on various industry-standard transport interfaces, the NVMe command set is also deployed across various transports. For example, SCSI has been supported on IP (iSCSI), Fibre Channel (SCSI_FCP, e.g., FCP), Serial Attached SCSI (SAS), and Parallel SCSI, among others.

Likewise, the NVMe command set has been implemented via PCIe, including U.2, as well as over fabrics (e.g., NVMe over Fabric, aka NVMeoF), most notably with TCP and RDMA, along with some NVMe over Fibre Channel. In addition to being a modern, streamlined command set, NVMe also supports more devices, queues, and sessions, and is optimized with host-server initiator and target-side stacks.
An example of how NVMe is being embraced from a software support standpoint is Microsoft’s production release of a new, re-architected, optimized I/O stack as part of Windows Server 2025 in December 2025. Recently, Microsoft released, as part of the latest vnext insiders preview build, an all-new NVMeoF (TCP & RDMA) initiator for testing and evaluation.

NVMe has proven itself as a command set using various transports for use as a target for various SSD and flash storage devices (e.g., back-end) in servers, clients, and storage systems, as well as a front-end alternative to SCSI_FCP (e.g., FCP or FC) and iSCSI among others on storage systems, and software-defined storage solutions. Also, keep in mind that while the SCSI command set dates back to at least the early 1990s for general-purpose access of Magnetic Tape, CD/DVD, HDD, and later SSD, NVMe developed in the last decade was created from the ground up for solid-state devices (SSDs), including flash, among other persistent storage mediums.

Why NVMeoF (TCP) vs iSCSI

It would be, and is, too easy to say the 2000s called and want iSCSI back; however, tongue-in-cheek, there is some reality to that. However, looking closer, iSCSI evolved in the early 2000s as a low-cost SCSI command set alternative to Fibre Channel (SCSI_FCP, e.g., FCP or FC), as well as the survivor of the block IP storage wars (e.g., iFCP, FCIP, iSCSI, etc.) for general-purpose IP-based SCSI storage access.

Fibre Channel-based SCSI (FCP) went on to dominate higher-performance fabric- or networked-storage, aka storage area networks (SANs), with iSCSI having some success in other areas alongside file-based IP storage access via NFS & SMB, among others.

Now that we are in the mid-2020s, the ecosystem around NVMe has matured and continues to do so, with full-stack NVMe solutions, hardware- and software-based initiators, native NVMe drivers, target-based storage systems, and storage devices using various transports. NVMe has evolved from direct-attached PCIe card slots to PCIe U.2 & U.3, M.2 and EDSFF slots, as well as NVMe over Fabrics including TCP and RDMA, among others.

The benefit of using NVMeoF vs iSCSI is lower latency, due to native NVMe commands over TCP being streamlined, vs older SCSI encapsulation and associated cumbersome driver stacks. Besides the streamlined drivers and transports, NVMeoF also provides massive parallelism and multi-threading to boost server storage I/O while leveraging today’s multi-core processors and offload capabilities.

The net result is more I/O operations supporting higher IOPs, bandwidth, and transactions per second, with lower latency and lower CPU usage, resulting in greater productivity and better economics for databases, virtualization, AI/ML, and other block-based workloads.

Who supports NVMe and NVMeoF?

Most host and client operating systems support NVMe or NVMeoF initiators (e.g., Linux, VMware, Microsoft Windows [in the current Insider build], among others). There are also hardware adapter-based NVMe initiators, such as those from Broadcom, Chelsio, and NVIDIA. There are software NVMe and NVMe target packages for many platforms, such as Linux-based NVMe, nvme-tcp, nvme-rdma, nvme-fc, nvmet, nvmet-tcp, and SPDK (Storage Performance Development Kit), among others.

Besides server- or client-side support, there are storage and server hardware chassis building blocks that support NVMe, as well as storage system solutions from Dell, HPE, Hitachi, NetApp, Pure, among others. Within large cloud service providers such as AWS, Microsoft Azure, and Google Cloud Platform (GCP), NVMe is also widely used across their platform ecosystems.

Why it’s time for NVMeoF to let iSCSI begin its slow retirement journey
This is not to say that iSCSI is dead or that it has no future; rather, NVMeoF and its ecosystem have evolved and are ready to do more. Likewise, the SCSI command set is not dead; it’s still used with SAS, Fibre Channel (FCP), iSCSI, etc. However, NVMe has evolved and is seeing broad adoption from server and client attachment to storage devices and storage systems over various transport interfaces and media (PCIe, U.2/U.3, M.2, EDSFF, Ethernet, and fabrics).

Storage systems have also evolved, adding front-end support for NVMeoF and back-end attachment of NVMe flash and SSDs. With the combination of server, client and other software platforms having or in public release review, new streamlined native NVMe drivers and I/O stacks along with NVMeoF initiators (or targets), now is the time to look at NVMeoF as a replacement or alternative to iSCSI, and perhaps even Fibre Channel in the future to support existing and legacy workloads as well as new and emerging application landscapes and their demanding server storage I/O and compute needs (e.g. AI among others).

Additional Resources Where to learn more

Microsoft Windows Server Insider Build NVMeoF Initiator (Blog post)
Installing new Insider’s build Windows Server vnext NVMeoF, host initiator (blog post)
NVMe Linux Driver and related info
Announcing Native NVMe in Windows Server 2025 (Microsoft Post)
Introducing the Windows NVMe-oF Initiator Preview in Windows Server Insiders Builds (Microsoft Post)
Windows Server Insider Builds (Microsoft Downloads)
ToE NVMeoF TCP Performance Line Boost Performance Reduce Costs (blog post)
TheNVMeplace.com (Various NVMe resources)
Azure Cloud Storage here (Microsoft Post)
Microsoft Azure Data Box (Blog Post)
Azure Elastic SAN from Cloud to On-Prem here (Blog Post)
Cloud and Software Defined Data Infrastructure topics here

Additional learning experiences along with common questions (and answers) pertaining to SCSI, Fibre Channel, NVMe, Cloud, Hardware, Software, Services, techniques, server, storage, I/O networking, data protection among other topics are found in my Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials (CRC Press) by Greg Schulz

What this all means

Now is a good time to look into NVMe and, specifically, NVMe over Fabric (NVMeoF) if you have not recently done so, to refresh or learn what it can do. Likewise, if you are using iSCSI or Fibre Channel-based SCSI_FCP (e.g., not FICON), now is a good time to review or refresh your understanding of NVMeoF and where it fits into future plans for block storage. For those who say block storage is dead and everything is object storage, file storage, or even table storage, that’s fine, as those are all companions to block storage in a balanced ecosystem, each with its own uses, roles, and fits.

Ok, nuff said.

Cheers Gs

Greg Schulz – Ten time Microsoft MVP Cloud and Data Center Management and Azure Storage, along with previous ten-time VMware vExpert. 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 2006-2026 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of UnlimitedIO LLC.

Posted on

Microsoft Windows Server Insider Build NVMeoF Initiator

Microsoft Windows Server Insider Build NVMeoF Initiator

In case you haven’t heard, in addition to recently announcing a new NVMe native driver for Windows Server 2025 back in December 2025, Microsoft has now introduced a new Windows NVMe over Fabrics (NVMeoF, aka NVMe-oF) Initiator Preview in the latest Windows Server Insiders Builds.

With the preview of this new NVMeoF initiator, Microsoft is moving closer to delivering enhanced server storage I/O performance by embracing NVMe as both a native storage target and a native NVMe host server initiator. While the new NVMe native storage stack driver was released for production back in December 2025, this new host initiator capability is available as a preview in the latest Windows Server insider build.

By adding a native NVMeoF initiator, Microsoft is enabling organizations to use shared networked block software-defined and SAN storage using the faster NVMe protocol, rather than traditional SCSI/iSCSI/SCSI_FCP(FC) protocol-based interfaces. In addition to supporting native direct-attached NVMe storage, Microsoft Windows servers (insider build) now support the streamlined NVMe server storage I/O stack over networks, including Ethernet and RDMA fabrics.

Additional Resources Where to learn more

ToE NVMeoF TCP Performance Line Boost Performance Reduce Costs (blog post)
NVMe Linux Driver and related info
Its Time for NVMeoF to let iSCSI begin its slow retirement journey (blog post)
Announcing Native NVMe in Windows Server 2025 (Microsoft Post)
Installing new Insider’s build Windows Server vnext NVMeoF, host initiator (blog post)
Introducing the Windows NVMe-oF Initiator Preview in Windows Server Insiders Builds (Microsoft Post)
Windows Server Insider Builds (Microsoft Downloads)
TheNVMeplace.com (Various NVMe resources)
Azure Cloud Storage here (Microsoft Post)
Microsoft Azure Data Box (Blog Post)
Azure Elastic SAN from Cloud to On-Prem here (Blog Post)
Cloud and Software Defined Data Infrastructure topics here

Additional learning experiences along with common questions (and answers) pertaining to Cloud, Hardware, Software, Services, techniques, server, storage, I/O networking, data protection among other topics are found in my Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials (CRC Press) by Greg Schulz

What this all means

While this new Windows Server host NVMe over Fabric (NVMeoF aka NVMe-oF) initiator for block storage is currently available as part of a Windows Server insiders build, it provides a means for organizations to begin testing native NVMe connectivity via Windows Servers to their existing SAN and other networked software defined storage servers via fabrics including Ethernet along with RDMA based. Learn more about this new initiator, along with how to try it out for yourself via this Microsoft blog post that includes instructions on how to setup your own NVMeoF storage target if you dont have one.

Ok, nuff said.

Cheers Gs

Greg Schulz – Ten time Microsoft MVP Cloud and Data Center Management and Azure Storage, along with previous ten-time VMware vExpert. 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 2006-2026 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of UnlimitedIO LLC.

Posted on

Microsoft Azure Elastic SAN from Cloud to On-Prem

What is Azure Elastic SAN

Azure Elastic SAN (AES) is a new (now GA) Azure Cloud native storage service that provides scalable, resilient, easy management with rapid provisioning, high performance, and cost-effective storage. AES (figure 1) supports many workloads and computing resources. Workloads that benefit from AES include tier 1 and tier 2, such as Mission Critical, Database, and VDI, among others traditionally relying upon consolidated Storage Area Network (SAN) shared storage.

Compute resources that can use AES, including bare metal (BM) physical machines (PM), virtual machines (VM), and containers, among others, using iSCSI for access. AES is accessible by computing resources and services within the Azure Cloud in various regions (check Azure Website for specific region availability) and from on-prem core and edge locations using iSCSI. The AES management experience and value proposition are similar to traditional hardware or software-defined shared SAN storage combined with Azure cloud-based management capabilities.

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 1 General Concept and Use of Azure Elastic SAN (AES)

While Microsoft Azure describes AES as a cloud-native storage solution, that does not mean that AES is only for containers and other cloud-native apps or DevOPS. Rather, AES has been built for and is native to the cloud (e.g., software-defined) that can be accessed by various compute and other resources (e.g., VMs, Containers, AKS, etc) using iSCSI.

How Azure Elastic SAN differs from other Azure Storage

AES differs from traditional Azure block storage (e.g., Azure Disks) in that the storage is independent of the host compute server (e.g., BM, PM, VM, containers). With AES, similar to a conventional software-defined or hardware-based shared SAN solution, storage is disaggregated from host servers for sharing and management using iSCSI for connectivity. By comparison, AES differs from traditional Azure VM-based storage typically associated with a given virtual machine in a DAS (Direct Attached Storage) type configuration. Likewise, similar to conventional on-prem environments, there is a mix of DAS and SAN, including some host servers that leverage both.

AES supports Azure VM, Azure Kubernetes Service (AKS), cloud-native, edge, and on-prem computing (BM, VM, etc.) via iSCSI. Support for Azure VMware Solution (AVS) is in preview; check the Microsoft Azure website for updates and new feature functionality enhancements.

Does this mean everything is moving to AES? Similar to traditional SANs, there are roles and needs for various storage options, including DAS, shared block, file, and object, among storage offerings. Likewise, Microsoft and Azure have expanded their storage offerings to include AES, DAS (azure disks, including Ultra, premium, and standard, among other options), append, block, and page blobs (objects), and files, including Azure file sync, tables, and Data Box, among other storage services.

Azure Elastic Storage Feature Highlights

AES feature highlights include, among others:

    • Management via Azure Portal and associated tools
    • Azure cloud-based shared scalable bock storage
    • Scalable capacity, low latency, and high performance (IOPs and throughput)
    • Space capacity-optimized without the need for data reduction
    • Accessible from within Azure cloud and from on-prem using iSCSI
    • Supports Azure compute  (VMs, Containers/AKS, Azure VMware Solution)
    • On-prem access via iSCSI from PM/BM, VM, and containers
    • Variable number of volumes and volume size per volume group
    • Flexible easy to use Azure cloud-based management
    • Encryption and network private endpoint security
    • Local (LRS) and Zone (ZRS) with replication resiliency
    • Volume snapshots and cluster support

Who is Azure Elastic SAN for

AES is for those who need cost-effective, shared, resilient, high capacity, high performance (IOPS, Bandwidth), and low latency block storage within Azure and from on-prem access. Others who can benefit from AES include those who need shared block storage for clustering app workloads, server and storage consolidation, and hybrid and migration. Another consideration is for those familiar with traditional hardware and software-defined SANs to facilitate hybrid and migration strategies.

How Azure Elastic SAN works

Azure Elastic SAN is a software-defined (cloud native if you prefer) block storage offering that presents a virtual SAN accessible within Azure Cloud and to on-prem core and edge locations currently via iSCSI. Using iSCSI, Azure VMs, Clusters, Containers, Azure VMware Solution among other compute and services, and on-prem BM/PM, VM, and containers, among others, can access AES storage volumes.

From the Azure Portal or associated tools (Azure CLI or PowerShell), create an AES SAN, giving it a 3 to 24-character name and specify storage capacity (base units with performance and any additional space capacity). Next, create a Volume Group, assigning it to a specific subscription and resource group (new or existing), then specify which Azure Region to use, type of redundancy (LRS or GRS), and Zone to use. LRS provides local redundancy, while ZRS provides enhanced zone resiliency, with highspeed synchronous resiliency without setting up multiple SAN systems and their associated replication configurations along with networking considerations (e.g., Azure takes care of that for you within their service).

The next step is to create volumes by specifying the volume name, volume group to use, volume size in GB, maximum IOPs, and bandwidth. Once you have made your AES volume group and volumes, you can create private endpoints, change security and access controls, and access the volumes from Azure or on-prem resources using iSCSI. Note that AES currently needs to be LRS (not ZRS) for clustered shared storage and that Key management includes using your keys with Azure key vault.

Using Azure Elastic SAN

Using AES is straightforward, and there are good easy to follow guides from Microsoft Azure, including the following:

The following images show what AES looks like from the Azure Portal, as well as from an Azure Windows Server VM and an onprem physical machine (e.g., Windows 10 laptop).

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 2 AES Azure Portal Big Picture

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 3 AES Volume Groups Portal View

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 4  AES Volumes Portal View

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 5 AES Volume Snapshot Views

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 6 AES Connected Volume Portal View

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 7 AES Volume iSCSI view from on-prem Windows Laptop

Microsoft Azure Elastic SAN from cloud to on-prem server storageioblog
Figure 8 AES iSCSI Volume attached to Azure VM

Azure Elastic SAN Cost Pricing

The cost of AES is elastic, depending on whether you scale capacity with performance (e.g., base unit) or add more space capacity. If you need more performance, add base unit capacity, increasing IOPS, bandwidth, and space. In other words, base capacity includes storage space and performance, which you can grow in various increments. Remember that AES storage resources get shared across volumes within a volume group.

Azure Elastic SAN is billed hourly based on a monthly per-capacity base unit rate, with a minimum of 1TB  provisioned capacity with minimum performance (e.g., 5,000 IOPs, 200MBps bandwidth). The base unit rate varies by region and type of redundancy, aka resiliency. For example, at the time of this writing, looking at US East, the Local Redundant Storage (LRS) base unit rate is 1TB with 5,000 IOPs and 200MBps bandwidth, costing $81.92 per unit per month.

The above example breaks down to a rate of $0.08 per GB per month, or $0.000110 per GB per hour (assumes 730 hours per month). An example of simply adding storage capacity without increasing base unit (e.g., performance) for US East is $61.44 per month. That works out to $0.06 per GB per month (no additional provisioned IOPs or Bandwidth) or $0.000083 per GB per hour.

Note that there are extra fees for Zone Redundant Storage (ZRS). Learn more about Azure Elastic SAN pricing here, as well as via a cost calculator here.

Azure Elastic SAN Performance

Performance for Azure Elastic SAN includes IOPs, Bandwidth, and Latency. AES IOPs get increased in increments of 5,000 per base TB. Thus, an AES with a base of 10TB would have 50,000 IOPs distributed (shared) across all of its volumes (e.g., volumes are not restricted). For example, if the base TB is increased from 10TB to 20TB, then the IOPs would increase from 50,000 to 100,000 IOPs.

On the other hand, if the base capacity (10TB) is not increased, only the storage capacity would increase from 10TB to 20TB, and the AES would have more capacity but still only have the 50,000 IOPs. AES bandwidth throughput increased by 200MBps per TB. For example, a 5TB AES would have 5 x 200MBps (1,000 MBps) throughput bandwidth shared across the volume groups volumes.

Note that while the performance gets shared across volumes, individual volume performance is determined by its capacity with a maximum of 80,000 IOPs and up to 1,024 MBps. Thus, to reach 80,000 IOPS and 1,024 MBps, an AES volume would have to be at least 107GB in space capacity. Also, note that the aggregate performance of all volumes cannot exceed the total of the AES. If you need more performance, then create another AES.

Will all VMs or compute resources see performance improvements with AES? Traditional Azure Disks associated with VMs have per-disk performance resource limits, including IOPs and Bandwidth. Likewise, VMs have storage limits based on their instance type and size, including the number of disks (HDD or SSD), performance (IOPS and bandwidth), and the number of CPUs and memory.

What this means is that an AES volume could have more performance than what a given VM is limited to. Refer to your VM instance sizing and configuration to determine its IOP and bandwidth limits; if needed, explore changing the size of your VM instance to leverage the performance of Azure Elastic SAN storage.

Additional Resources Where to learn more

The following links are additional resources to learn about Microsoft Azure Elastic SAN and related data infrastructures and tradecraft topics.

Windows Server Insider Build new NVMeoF initiator (Blog Post)
Azure AKS Storage Concepts 
Azure Elastic SAN (AES) Documentation and Deployment Guides
Azure Elastic SAN Microsoft Blog
Azure Elastic SAN Overview
Azure Elastic SAN Performance topics
Azure Elastic SAN Pricing calculator
Azure Products by Region (see where AES is currently available)
Azure Storage Offerings 
Azure Virtual Machine (VM) sizes
Azure Virtual Machine (VM) types
Azure Elastic SAN General Pricing
Azure Storage redundancy 
Azure Service Level Agreements (SLA) 
StorageIOBlog.com Data Box Family 
StorageIOBlog.com Data Box Review
StorageIOBlog.com Data Box Test Drive 
StorageIOblog.com Microsoft Hyper-V Alive Enhanced with Win Server 2025
StorageIOblog.com If NVMe is the answer, what are the questions?
StorageIOblog.com NVMe Primer (or refresh)
RTO Context Matters (Blog Post)

Additional learning experiences along with common questions (and answers), are found in my Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials (CRC Press) by Greg Schulz

What this all means

Azure Elastic SAN (AES) is a new and now generally available shared block storage offering that is accessible using iSCSI from within Azure Cloud and on-prem environments. Even with iSCSI, AES is relatively easy to set up and use for shared storage, mainly if you are used to or currently working with hardware or software-defined SAN storage solutions.

With NVMe over TCP fabrics gaining industry and customer traction, I’m hoping for Microsoft to adding that in the future. Currently, AES supports LRS and ZRS for redundancy, and an excellent future enhancement would be to add Geo Redundant Storage (GRS) capabilities for those who need it.

I like the option of elastic shared storage regarding performance, availability, capacity, and economic costs (PACE). Suppose you understand the value proposition of evolving from dedicated DAS to shared SAN (independent of the underlying fabric network); or are currently using some form of on-prem shared block storage. In that case, you will find AES familiar and easy to use. Granted, AES is not a solution for everything as there are roles for other block storage, including DAS such as Azure disks and VMs within Azure, along with on-prem DAS, as well as file, object, and blobs, tables, among others.

Wrap up

The notion that all cloud storage must be objects or blobs is tied those who only need, provide, or prefer those solutions. The reality is that everything is not the same. Thus, there is a need for various storage mediums, devices, tiers, access, and types of services. Microsoft and Azure have done an excellent job of providing. I like what Microsoft Azure is doing with Azure Elastic SAN.

Ok, nuff said, for now.

Cheers Gs

Greg Schulz – Nine time Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2018. 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 2006-2026 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of UnlimitedIO LLC.

Posted on

Microsoft Hyper-V Is Alive Enhanced With Windows Server 2025

Yes, you read that correctly, Microsoft Hyper-V is alive and enhanced with Windows Server 2025, formerly Windows Server v.Next server. Note that  Windows Server 2025 preview build is just a preview available for download testing as of this time.

What about Myth Hyper-V is discontinued?

Despite recent FUD (fear, uncertainty, doubt), misinformation, and fake news, Microsoft Hyper-V is not dead. Nor has Hyper-V been discontinued, as some claim. Some Hyper-V FUD is tied to customers and partners of VMware following Broadcom’s acquisition of VMware looking for alternatives. More on Broadcom and VMware here, here, here, here, and here.

As a result of Broadcom’s VMware acquisition and challenges for partners and customers (see links above), organizations are doing due diligence, looking for replacement or alternatives. In addition, some vendors are leveraging the current VMware challenges to try and position themselves as the best hypervisor virtualization safe harbor for customers. Thus some vendors, their partners, influencers and amplifiers are using FUD to keep prospects from looking at or considering Hyper-V.

Virtual FUD (vFUD)

First, let’s shut down some Virtual FUD (vFUD). As mentioned above, some are claiming that Microsoft has discontinued Hyper-V. Specifically, the vFUD centers on Microsoft terminating a specific license SKU (e.g., the free Hyper-V Server 2019 SKU). For those unfamiliar with the discontinued SKU (Hyper-V Server 2019), it’s a headless (no desktop GUI) version of Windows Server  running Hyper-V VMs, nothing more, nothing less.

Does that mean the Hyper-V technology is discontinued? No.

Does that mean Windows Server and Hyper-V are discontinued? No.

Microsoft is terminating a particular stripped-down Windows Server version SKU (e.g. Hyper-V Server 2019) and not the underlying technology, including Windows Server and Hyper-V.

To repeat, a specific SKU or distribution (Hyper-V Server 2019) has been discontinued not Hyper-V. Meanwhile, other distributions of Windows Server with Hyper-V continue to be supported and enhanced, including the upcoming Windows Server 2025 and Server 2022, among others.

On the other hand, there is also some old vFUD going back many years, or a decade, when some last experienced using, trying, or looking at Hyper-V. For example, the last look at Hyper-V might been in the Server 2016 or before era.

If you are a vendor or influencer throwing vFUD around, at least get some new vFUD and use it in new ways. Better yet, up your game and marketing so you don’t rely on old vFUD. Likewise, if you are a vendor partner and have not extended your software or service support for Hyper-V, now is a good time to do so.

Watch out for falling into the vFUD trap thinking Hyper-V is dead and thus miss out on new revenue streams. At a minimum, take a look at current and upcoming enhancements for Hyper-V doing your due diligence instead of working off of old vFUD.

Where is Hyper-V being used?

From on-site (aka on-premises, on-premises, on-prem) and edge on Windows Servers standalone and clustered, to Azure Stack HCI. From Azure, and other Microsoft platforms or services to Windows Desktops, as well as home labs, among many other scenarios.

Do I use Hyper-V? Yes, when I  retired from the vExpert program after ten years. I moved all of my workloads from VMware environment to Hyper-V including *nix, containers and Windows VMs, on-site and on Azure Cloud.

How Hyper-V Is Alive Enhanced With Windows Server 2025

Is Hyper-V Alive Enhanced With Windows Server 2025?  Yup.

Formerly known as Windows Server v.Next, Microsoft announced the Windows Server 2025 preview build on January 26, 2024 (you can get the bits here). Note that Microsoft uses Windows Server v.Next as a generic placeholder for next-generation Windows Server technology.

A reminder that the cadence of Windows Server Long Term Serving Channels (LTSC) versions has been about three years (2012R2, 2016, 2019, 2022, now 2025), along with interim updates.

What’s enhanced with Hyper-V and Windows Server 2025

    • Hot patching of running server (requires Azure Arc management) with almost instant implementations and no reboot for physical, virtual, and cloud-based Windows Servers.
    • Scaling of even more compute processors and RAM for VMs.
    • Server Storage I/O performance updates, including NVMe optimizations.
    • Active Directory (AD) improvements for scaling, security, and performance.
    • There are enhancements to storage replica and clustering capabilities.
    • Hyper-V GPU partition and pools, including migration of VMs using GPUs.

More Enhancements for Hyper-V and Windows Server 2025

Active Directory (AD)

Enhanced performance using all CPUs in a process group up to 64 cores to support scaling and faster processing. LDAP for TLS 1.3, Kerberos support for AES SHA 256 / 384, new AD functional levels, local KDC, improved replication priority, NTLM retirement, local Kerberos, and other security hardening. In addition, 64-bit Long value IDs (LIDs) are supported along with a new database schema using 32K pages vs the previous 8K pages. You will need to upgrade forest-wide across domain controllers to leverage the new larger page sizes (at least Server 2016 or later). Note that there is also backward compatibility using 8K pages until all ADs are upgraded.

Storage, HA, and Clustering

Windows Server continues to offer flexible options for storage how you want or need to use it, from traditional direct attached storage (DAS) to Storage Area Networks (SAN), to Storage Spaces Direct (S2D) software-defined, including NVMe, NVMe over Fabrics (NVMeoF), SAS, Fibre Channel, iSCSI along with file attached storage. Some other storage and HA enhancements include Storage Replica performance for logging and compression and stretch S2D multi-site optimization.

Failover Cluster enhancements include AD-less clusters, cert-based VM live migration for the edge, cluster-aware updating reliability, and performance improvements. ReFS enhancements include dedupe and compression optimizations.

Other NVMe enhancements include optimization to boost performance while reducing CPU overhead, for example, going from 1.1M IOPS to 1.86M IOPS, and then with a new native NVMe driver (to be added), from 1.1M IOPs to 2.1M IOPs. These performance optimizations will be interesting to look at closer, including baseline configuration, number and type of devices used, and other considerations.

Compute, Hyper-V, and Containers

Microsoft has added and enhanced various Compute, Hyper-V, and Container functionality with Server 2025, including supporting larger configurations and more flexibility with GPUs. There are app compatibility improvements for containers that will be interesting to see and hear more details about besides just Nano (the ultra slimmed-down Windows container).

Hyper-V

Microsoft extensively uses Hyper-V technology across different platforms, including Azure, Windows Servers, and Desktops. In addition, Hyper-V is commonly found across various customer and partner deployments on Windows Servers, Desktops, Azure Stack HCI, running on other clouds, and virtualization (nested). While Microsoft effectively leverages Hyper-V and continues to enhance it, its marketing has not effectively told and amplified the business benefit and value, including where and how Hyper-V is deployed.

Hyper-V with Server 2025 includes discrete device assignment to VM (e.g., resources dedicated to VMs). However, dedicating a device like a GPU to a VM prevents resource sharing, failover cluster, or live migration. On the other hand, Server 2025 Hyper-V supports GPU-P (GPU Partitioning), enabling GPU(s) to be shared across multiple VMs. GPUs can be partitioned and assigned to VMs, with GPUs and GPU partitioning enabled across various hosts.

In addition to partitioning, GPUs can be placed into GPU pools for HA. Live migration and cluster failover (requires PCIe SR-IOV), AMD Lilan or later, Intel Sapphire Rapids, among other requirements, can be done. Another enhancement is Dynamic Processor Compatibility, which allows mixed processor generations to be used across VMs and then masks out functionalities that are not common across processors. Other enhancements include optimized UEFI, secure boot, TPM , and hot add and removal of NICs.

Networking

Network ATC provides intent-based deployments where you specify desired outcomes or states, and the configuration is optimized for what you want to do. Network HUD enables always-on monitoring and network remediation. Software Defined Network (SDN) optimization for transparent multi-site L2 and L3 connectivity and improved SDN gateway performance enhancements.

SMB over QUIC leverages TLS 1.3 security to streamline local, mobile, and remote networking while enhancing security with configuration from the server or client. In addition, there is an option to turn off SMB NTLM at the SMB level, along with controls on which versions of SMB to allow or refuse. Also being added is a brute force attack limiter that slows down SMB authentication attacks.

Management, Upgrades, General user Experience

The upgrade process moving forward with Windows Server 2025 is intended to be seamless and less disruptive. These enhancements include hot patching and flighting (e.g., LTSC Windows server upgrades similar to how you get regular updates). For hybrid management, an easier-to-use wizard to enable Azure Arc is planned. For flexibility, if present, WiFi networking and Bluetooth devices are automatically enabled with Windows Server 2025 focused on edge and remote deployment scenarios.

Also new is an optional subscription-based licensing model for Windows Server 2025 while retaining the existing perpetual use. Let me repeat that so as not to create new vFUD, you can still license Windows Server (and thus Hyper-V) using traditional perpetual models and SKUs.

Additional Resources Where to learn more

The following links are additional resources to learn about Windows Server, Server 2025, Hyper-V, and related data infrastructures and tradecraft topics.

ToE NVMeoF TCP Boost Performance Reduce Costs (blog post)
NVMe Linux Driver and related info
Announcing Native NVMe in Windows Server 2025 (Microsoft Post)
Introducing the Windows NVMe-oF Initiator Preview in Windows Server Insiders Builds (Microsoft Post)
Its Time for NVMeoF to let iSCSI begin its slow retirement journey (blog post)
What’s New in Windows Server v.Next video from Microsoft Ignite (11/17/23)
Microsoft Windows Server 2025 Whats New
Microsoft Windows Server 2025 Preview Build Download
Microsoft Windows Server 2025 Preview Build Download (site)
Microsoft Evaluation Center (various downloads for trial)
Microsoft Eval Center Windows Server 2022 download
Microsoft Hyper-V on Windows Information
Microsoft Hyper-V on Windows Server Information
Microsoft Hyper-V on Windows Desktop (e.g., Win10)
Microsoft Windows Server Release Information
Microsoft Hyper-V Server 2019
Microsoft Azure Virtual Machines Trial
Microsoft Azure Elastic SAN
If NVMe is the answer, what are the questions?
NVMe Primer (or refresh), The NVMe Place.

Additional learning experiences along with common questions (and answers), are found in my Software Defined Data Infrastructure Essentials book.

Software Defined Data Infrastructure Essentials Book SDDC

What this all means

Hyper-V is very much alive, and being enhanced. Hyper-V is being used from Microsoft Azure to Windows Server and other platforms at scale, and in smaller environments.

If you are looking for alternatives to VMware or simply exploring virtualization options, do your due diligence and check out Hyper-V. Hyper-V may or may not be what you want; however, is it what you need? Looking at Hyper-V now and upcoming enhancements also positions you when asked by management if you have done your due  diligence vs relying on vFUD.

Do a quick Proof of Concept, spin up a lab, and check out currently available Hyper-V. For example, on Server 2022 or 2025 preview, to get a feel for what is there to meet your needs and wants. Download the bits and get some hands on time with Hyper-V and Windows Server 2025.

Wrap up

Hyper-V is alive and enhanced with Windows Server 2025 and other releases.

Ok, nuff said, for now.

Cheers Gs

Greg Schulz – Nine time Microsoft MVP Cloud and Data Center Management, VMware vExpert 2010-2018. 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 2006-2026 Server StorageIO and UnlimitedIO. All Rights Reserved. StorageIO is a registered Trade Mark (TM) of UnlimitedIO LLC.