Working out processor utilzation of current ESX environment. Please help!

[dilpz]

Hi Guys,

Ive been given the task of replacing our end of life ESX 3.0.2 servers with new IBM kit and am currently going through speccing out what we need. Currently, we have 2 IBM hosts each containing 8 dual-core 3.0ghz xeons in them (total of 32 cores between the 2) which need to be replaced. We also have another smaller server with 4 dual core 3.0ghz xeons in it which still has 2 years to run. I need to purchase an IBM solution which will provide us with enough processing power to cope with usage at the minute.

Currently, I have measured what we are using now by looking at the performance tab within Virtualcenter for each host and looking at the average CPU usage in mhz over the past month. Using this information I worked out that e.g. for one of the boxes, the average use in MHZ over the past month was 10114.4mhz and therefore on average 4 of the 16 3.0Ghz cores were being utilized over the past month. Is this correct? Is this the best way to look at this or can anyone recommend a better way?

Any advice will be much appreciated.

[Texiwill]

Hello,

Using VirtualCenter is definitely a way to go, and where I would start. I would not look at the total CPU utilization but the per core utilization. You may find that each core is 50% utilized. Which may be where you want everything to be. If so then you are set and should upgrade accordingly. However, if you find one core overutilized (greater than 80%) then something may be skewed and you need to consider a different option.

Those are some big boxes, but you may be able to switch them easily with 4 quad cores per box..... Or slightly cheaper go with 2x systems but 2 quad cores per server... Depends on how much you really need to spend and how much redundancy you wish to add to your system.

CPU is not everything, I would look at your networking, etc. This is a good time to fix any issues you currently have with the infrastructure whether they be Performance, disk layout, or security.

Best regards,

Edward L. Haletky

VMware Communities User Moderator

====

Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education. CIO Virtualization Blog: http://www.cio.com/blog/index/topic/168354, As well as the Virtualization Wiki at http://www.astroarch.com/wiki/index.php/Virtualization

[FredPeterson]

How many VM's do you have? Or in another light how many vCPU's are in use?

I would probably go with 3 independent x3850M2's with 4 Quad Core 2.4's or 2.93's (totally dependent upon expectations of your customers) and appropriately sized memory. 48 cores total, but spread equally 16 cores on each host. Gives you more flexibility in uptime because you can more easily take one host down and not overwhelm the others when you enter maintenance mode and DRS kicks in. You might even be able to squeeze down to just 2 x3850M2's - but then thats a lot of VM's that crash and burn should the host go down.

Statistics like percent ready as well as your current "consolidation ratio" - how many VM's / Core - can be more important then CPU utilization.

Rolled up statistics in VirtualCenter are going to be averages. So that 10GHz average could be the result of a few crazy spikes to 100% CPU usage. High water marks can be more important - how critical is horsepower when its really needed.

[Texiwill]

Hello,

For those who do not have VC, you can get the same information from vmktree and unnoc.

Best regards,

Edward L. Haletky

VMware Communities User Moderator

====

Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education. CIO Virtualization Blog: http://www.cio.com/blog/index/topic/168354, As well as the Virtualization Wiki at http://www.astroarch.com/wiki/index.php/Virtualization

[dilpz]

Hi Guys,

Thank you all for your replies, some really helpful advice there. In answer to your questions, we currently run 56vcpu's on 42VMs that are spread across all 3 hosts. I need to keep the same amount of processing power but, as we are being encouraged to consolidate to a single datacenter, over the next 3 years (for which time this solution should last) some of these VMs will disappear. One of you mentioned looking at the per core utilisation. How would you go about doing this? Is it within VC?

Thanks

Dilpz

[dilpz]

Hi Guys,

I have just used ESXtop to look at the avg per physical CPU utilisation in real time for each of my boxes. For the half hour I looked, all hosts fluctuated between 20-40 %. Do you think this is a good measure to base any proposed solutions on?

[jitendrakmr]

looks like what you have in current can hold many more VMs so there is still scope of adding more VMs to you existing Vmware farm. Why are you planning to buy new hardware?

anyway If you are looking for same processing power then you probably need not to go into more details of checking total CPU utilization because the current utilization shows that even servers with lower configuration can run the current VMs in environment but in that case there will be no scope of adding more VMs to environment and yes as other said if some of cores are heavily utilized then you probably look for solution having more number of cores than the current number of heavily utilized cores.

Also as a thumb rule monitor peak CPU utlization of current VMs and the solution which you are going to buy should have enough room to hold all Vms while they are running on their peak CPU utilization.

Thanks.

Jitendra Kumar

MCSE 2003, VCP, CCNA, ITIL Foundation