VM Capacity Planning

[cmputrnrd4lif]

I'm currently planning my company's move to virtualization, but am unsure how best to gauge current server utilization. I've messed around with Windows perfmon a bit, but I'm not getting data that seems to be very useful to me. Are there any free options or utilities that can be purchased?

I'm aware that VM resellers have a Capacity Planner tool, but what if I don't want to have a VM partner come in to do it?

I need to be able to successfully calculate our current usage so I don't buy new equipment that isn't going to handle the load. It'll also be nice to know how much room we have to expand.

Thanks.

 

[dx2]

Any body have any white papers handy on this topic? Particularly in capacity planning for a QA/Lab environment where kicking out new servers needs to be able to be done in a quick logical fasion to make the most of available hardware.

 

[Haven]

We are planning on doing a virtualization of some servers that we are currently renting and moving the VMs into our own hardware. Since we were looking to buy VMWare ESX we got the Capacity planning for free. Though I am looking to spend around $200K with the vendor for servers, software, SAN, etc. So that might be why they did it for free for us.

 

[Berg0]

just sign up for the 60 day trial of ESX, and download capacity IQ ovf template? shouldn't need to have anyone come in to do that.

 

[crazyfinx]

just sign up for the 60 day trial of ESX, and download capacity IQ ovf template? shouldn't need to have anyone come in to do that.

hehe capacity IQ is not supported with vSphere hosts at the moment, only esxi/esx 3.x or older, esx/esxi 4.x not supported, try using vkernel's capacity analyzer trial or free modeler

 

[subflava]

I'd also be curious about any good tools someone has actually used/found. I've only planned VM capacity manually with various tools (mostly graphing past performance either with SNMP monitors like Cacti or Perfmon). It seems pretty complicated for a "free" type of tool to be able to accomplish with any sort of accuracy. There's basically 3 main potential bottlenecks I see with virtualization; CPU usage, memory usage, and disk I/O. Memory usage is probably easiest to calculate as you can pretty much add up all the memory being used on your servers (actual usage, not what's in the box) which usually reaches a steady state on most servers. If there's any spikes/peaks for periodic processes, you'd have to account for those.

CPU usage is a little more complicated, but pretty much the same idea. It's just that CPU usage can be a lot more bursty, but even still there is usually a predictable pattern on most servers. You'll need to graph this for at least a week with a tool to get a good idea if you're not already doing it. You'll have to do some analysis on the average usage/95th percentile/where peaks coincide, etc.

The I/O analysis is probably the hardest as usually most people offload that to a SAN or some sort of shared/external storage device. Depending on your requirements, that could get complicated on it's own although for most small offices people generally just attach a big SAN (as big as they can afford anyways) on the network and hope it works well enough. Your storage vendor would be more than happy to work with you on that.

Anyways, back to the original question about a planning tool...I'm not sure I'd trust a tool to be able to accurately judge all this information for me. However, I'm sure it could give a good rough idea if that's all your after.