VMware Cloud Community
Konsolidator
Contributor
Contributor

virtualization criterias

Hi everyone,

Engaged in a huge X86 server consolidation program, one of our main concern is about the virtualisation cirterias : more clearly, what can be virtualized easely, without other limitation than physical one (CPU usage, IO, Ram, ...) typically, Tier 1 servers. What are criterias for application servers or WAS and what is not natively candidate to be virtualized. And the same for Tier 3 and dataservers. I suppose some of you had made this exercice, if you have mething on this, it could help us.

Thanks for awnser, if you have some ideas.

Reply
0 Kudos
7 Replies
MartijnLo
Enthusiast
Enthusiast

I am assuming that you have the time to test properly etc.

In that case the one criteria you could use is "we will virtualize everything unless it can be proven to not work while virtualized".

Reply
0 Kudos
dwang1092
Contributor
Contributor

x

Reply
0 Kudos
aenagy
Hot Shot
Hot Shot

I second that. In our environment the choices of platforms, in decending preference, are: (1) virtualization (in our case VI3), (2) blade, and (3) rack mount server. This applies to all production and non-production (dev, test, QA, production support). The only way an application ends up not going virtual is if testing indicates that performance is unacceptable or their is some specific hardware dependency, e.g. fax server.

We encountered a vendor that told us that thier application is not supported in a virtualized environment due to performance. Actually, I think this was FUD on thier part, but I can't prove it. After some research and testing we found that the performance of the application on an ESX Server 3.0.1 virtual machine was at most 5% slower than the same application on a physical machine. What we found from the research and testing was: (1) coding of the workflow in the application was very inefficient and was artificially slowing down the user's experience, (2) the first implementation of the application in our environment was on ESX Server 2.5, once we migrated the virtual machines to ESX Server 3.0.1 performance increased dramatically (25%), and (3) moving the application's database (SQL Server 2005) from a virtual machine to a physical machine also increased the performance dramatically (50%). The point is that without research and testing the internal customer and the vendor would have insisted on physical machines. As we go through some lease roll-overs for our server hardware everything will be virtualized with a few execeptions: (1) production Exchange, (2) production SQL Server, (3) at lease one domain controller in each domain all other domain controllers will be virtualized, and (4) any server with a hardware dependency.

Two performance metrics that I watched closely when testing were "CPU Ready" of the virtual machine, and "Processor > % Privileged Time" in perfmon. I have been told by our VMware SE that VMware has a performance team that will help customers with performance related issues. You would likely need some sort of support agreement with VMware or one of thier partners such as HP or IBM to open a case before engaging this team.

Hope that helps.

Reply
0 Kudos
Konsolidator
Contributor
Contributor

Hi,

Thanks for your awnser. It is the way they follow to make the P2V migration. But I am not so satisfied with this mehtod, because we have a huge number of servers (+ 30 000) and it takes a long time. We would like to have an industrial real process for this.

Perhaps we could in a first time make a selection of what is natively virtualizable, without specific study (generaly Tier 1), for what we have to study more deeply, and what are the applications we can't virtualize. I have some experiences from other major companies who made this, and for example, they did'nt put the Tier 3 (databases) in the scope to be virtualized. This was my concern in the question, but I am effraid that not a lot of companies have made this reflection.

For my opinion, 2 kinds of criterias have to be defined :

Technical cirterias, based on CPU, IO, Network and RAM usage, but also on hardware specific components used (X25 cards, ...)

Funtionnal criterias : based on application type (OLTP, Batch, ...), SLA, I/O profiles, ... this is because virtualization technology is just a tool to consolidate applications on the same infrastructure.

Best regards

Eric Stern

Orange Corp - FT - ROSI-ITNPS/IT Governance

Tel : + 33 1 58 88 63 74

eric.stern@orange-ftgroup.com

Reply
0 Kudos
Konsolidator
Contributor
Contributor

Hi,

Thank you for awnsering me !

Yes, it helps me, I just check this with one of our expert.

Regards

Eric Stern

Orange Corp - FT - ROSI-ITNPS/IT Governance

Tel : + 33 1 58 88 63 74

eric.stern@orange-ftgroup.com

Reply
0 Kudos
MartijnLo
Enthusiast
Enthusiast

I would scope the virtualization candidates through a Capacity Planner or PowerRecon analysis. Low CPU utilization Windows and Linux machines are primary candidates. Also test and development environments would be logical to virtualize first. Anything that has low service level requirements attached to them so it can be done during regular office hours is easy to pick up and virtualize and you can make rapid progress with those kinds of servers. This in turn can be helpful in proving your progress and realization of the benefits to higher echelons within the company.

As far as industrializing the virtualization process: that can be difficult because especially business servers have business users and owners attached to them and they need to be informed about the migration and (rightfully or not) they are bound to want to influence your process. Stakeholder management is very important in that regard.

Reply
0 Kudos
aenagy
Hot Shot
Hot Shot

This might be of interest too:

How to run 16,000 Exchange mailboxes on ESX

http://vmetc.com/2008/02/27/how-to-run-16000-exchange-mailboxes-on-esx/

Reply
0 Kudos