Proving ESXi Performance

[AdamW201110141]

Hey there,

I just took a new job a few months back as a Virtualization Engineer. Unfortunately, the last guy made a mess of the current environment and the company has kind of lost confidence in anything virtualization (and rightfully so. the last environment ran horrible.) Mostly due to performance and ease of deploying VMs. I have been working to fix the issues, but now am at the point where I need to prove that the Guests and ESXi 4.1 hosts are a good option over Physical machines. Aside from the cost savings and DR capabilities, I need to come up with some performance metrics compared to what we can do on physical servers.

So my question:  What good tools are out there that will help with looking at guest performance, network perfromance, and possibly disk. I'm trying to get to where I can show an application or DB server can perform at a very similar level to a physical box with the same specs.

Windows 2003/2008

ESXi 4.1 U1, HP blade and Cisco UCS blade hardware.

Thanks in advance.

-Adam

[DSTAVERT]

You don't mention storage which is where most performance problems begin.

Performance Best Practices.

http://www.vmware.com/resources/techresources/10161

Virtualizing Enterprise Applications

http://www.vmware.com/solutions/business-critical-apps/index.html

[AdamW201110141]

Thanks. I've seem many of the docs from the links and am tweeking things still a bit.

Storage is definately the most improved part and what I believe to be the lions share cause of the issues we had. I've got a working solution in place until they can upgrade the backed storage to a more robust solution. And it's working nicely. Now just to prove performance.

I think really what they are looking for is App level performance:

     Data thruput over the network

     Data/Read Write speeds to the storage

     DB Performance

     Recognized app perfromance

          The ability to run the same function in about the same time - Physical vs. Virtual

I can already see a performance gain, but their not goign to take "It just runs better" comment from me.

Looking for tools that I can run on a physical node and virtual node, run the same process, and compare the results. You know pretty graphs, or tables that I can get saw raw numbers from.

-Adam

[DSTAVERT]

Have a look through some of Scott Drummonds documents. http://communities.vmware.com/people/drummonds?view=documents He was VMware's performance troubleshooter.

[mcowger]

Dont use anything thats not end-user focused.  Who cares if CPU utilization or DB read speeds vary?  As long as the end users get their results back in the same 'N' milliseconds, thats all that matters.

Test on physical, then make sure your virtual gets similar resources and test there.

There are a number of white papers out there showing, for example, that exchange runs better as a set of VMs on a host than the host running natively.