Should everything, that is all servers, storage and I/O along with facilities, be virtualized?
The answer not surprisingly should be it depends!
Denny Cherry (aka Mrdenny) over at ITKE did a great recent post about applications not being virtualized, particularly databases. In general some of the points or themes we are on the same or similar page, while on others we slightly differ, not by very much.
Unfortunately consolidation is commonly misunderstood to be the sole function or value proposition of server virtualization given its first wave focus. I agree that not all applications or servers should be consolidated (note that I did not say virtualized).
From a consolidation standpoint, the emphasis is often on boosting resource use to cut physical hardware and management costs by boosting the number of virtual machines (VMs) per physical machine (PMs). Ironically, while VMs using VMware, Microsoft HyperV, Citrix/Xen among others can leverage a common gold image for cloning or rapid provisioning, there are still separate operating system instances and applications that need to be managed for each VM.
Sure, VM tools from the hypervisor along with 3rd party vendors help with these tasks as well as storage vendor tools including dedupe and thin provisioning help to cut the data footprint impact of these multiple images. However, there are still multiple images to manage providing a future opportunity for further cost and management reduction (more on that in a different post).
Getting back on track:
Some reasons that all servers or applications cannot be consolidated include among others:
- Performance, response time, latency and Quality of Service (QoS)
- Security requirements including keeping customers or applications separate
- Vendor support of software on virtual or consolidated servers
- Financial where different departments own hardware or software
- Internal political or organizational barriers and turf wars
On the other hand, for those that see virtualization as enabling agility and flexibility, that is life beyond consolidation, there are many deployment opportunities for virtualization (note that I did not say consolidation). For some environments and applications, the emphasis can be on performance, quality of service (QoS) and other service characteristics where the ratio of VMs to PMs will be much lower, if not one to one. This is where Mrdenny and me are essentially on the same page, perhaps saying it different with plenty of caveats and clarification needed of course.
My view is that in life beyond consolidation, many more servers or applications can be virtualized than might be otherwise hosted by VMs (note that I did not say consolidated). For example, instead of a high number or ratio of VMs to PMs, a lower number and for some workloads or applications, even one VM to PM can be leveraged with a focus beyond basic CPU use.
Yes you read that correctly, I said why not configure some VMs on a one to one PM basis!
Here’s the premise, todays current wave or focus is around maximizing the number of VMs and/or the reduction of physical machines to cut capital and operating costs for under-utilized applications and servers, thus the move to stuff as many VMs into/onto a PM as possible.
However, for those applications that cannot be consolidated as outlined above, there is still a benefit of having a VM dedicated to a PM. For example, by dedicating a PM (blade, server or perhaps core) allows performance and QoS aims to be meet while still providing the ability for operational and infrastructure resource management (IRM), DCIM or ITSM flexibility and agility.
Meanwhile during busy periods, the application such as a database server could have its own PM, yet during off-hours, some over VM could be moved onto that PM for backup or other IRM/DCIM/ITSM activities. Likewise, by having the VM under the database with a dedicated PM, the application could be moved proactively for maintenance or in a clustered HA scenario support BC/DR.
What can and should be done?
First and foremost, decide how VMs is the right number to divide per PM for your environment and different applications to meet your particular requirements and business needs.
Identify various VM to PM ratios to align with different application service requirements. For example, some applications may run on virtual environments with a higher number of VMs to PMs, others with a lower number of VMs to PMs and some with a one VM to PM allocation.
Certainly there will be for different reasons the need to keep some applications on a direct PM without introducing a hypervisors and VM, however many applications and servers can benefit from virtualization (again note, I did not say consolation) for agility, flexibility, BC/DR, HA and ease of IRM assuming the costs work in your favor.
Additional general to do or action items include among others:
- Look beyond CPU use also factoring in memory and I/O performance
- Keep response time or latency in perspective as part of performance
- More and fast memory are important for VMs as well as for applications including databases
- High utilization may not show high hit rates or effectiveness of resource usage
- Fast servers need fast memory, fast I/O and fast storage systems
- Establish tiers of virtual and physical servers to meet different service requirements
- See efficiency and optimization as more than simply driving up utilization to cut costs
- Productivity and improved QoS are also tenants of an efficient and optimized environment
These are themes among others that are covered in chapters 3 (What Defines a Next-Generation and Virtual Data Center?), 4 (IT Infrastructure Resource Management), 5 (Measurement, Metrics, and Management of IT Resources), as well as 7 (Servers—Physical, Virtual, and Software) in my book “The Green and Virtual Data Center (CRC) that you can learn more about here.
Ok, nuff said.
All Comments, (C) and (TM) belong to their owners/posters, Other content (C) Copyright 2006-2013 StorageIO All Rights Reserved