Having had a lot of experience with GSX and VMWare Server (we have about a 100 VMs running on six VMWare servers), we were enthusiastic to deploy ESX3.
We first took a Dell PE1850 server with dual Xeon and 4GB RAM and moved a few VMs from VMWare Server to ESX.
However, they ran so much slower than they had under VMWare Server that we decided we needed more hardware to run ESX. We switched to a Dell PE1950 with dual quad-core CPUs and 8GB RAM and tried again.
Regardless of how much CPU resources we allocated to the VMs, we could not achieve performance anywhere near that which we were seeing under VMWare server (we were measuring the performance of the Windows 2000 server guests with a test tool which tested the performance of our web application).
Both servers were on the HCL for VMWare Virtual Infrastructure 3 and both installed VI3 without problems.
We love the idea of fine-grained control over the resources allocated to the guests, but not if they run at half the speed of the VMs under VMWare Server.
Q1. Are there any published benchmarks of relative performance of identical guests under VMWare Server and VI3? What kind of performance difference should I be expecting?
Q2. What kind of server do I need to buy to make VI3 run so well that I'm not tempted to replace it with a bunch of smaller servers running faster?
I know that the real USP of VI3 is enterprise manageability etc, but its hard to recommend my customers to pay for a significant drop in performance relative to the free version...
how exactly are u measuing performance, ESX should always run faster than a host based solution, i have used GSX and ESX, the performace ESX has over GSX is massive.
Please provide more info about your config, something must be wrong!!
According to VMware ESX is roughly twice as efficient as VMware Server, with VMware Server providing 2-4 VMs per core to ESX's 4-8 VMs per core.
I was reading a doc on this last week but I can't seem to find it now, if I do I'll post it up for you.
As for your server, ESX should run just fine on a dual Xeon 4GB box, let alone a dual quad-core with 8GB, so I'm thinking you have a problem somewhere with your ESX install.
Thanks a lot for all the feedback everyone.
Regarding epping's question - how did we measure performance? We measured the performance of 4 web-applications each running as a Windows 2000 Server guest. We used a web client test tool which measures the time required by the servers to process a set of requests generated by the client. It should provide a very objective measurement of the performance of the servers.
Anyway, it seems like we must have had some problem with our specific configuration. We installed VI3 onto the brand-new Dell 1950 and used VMWare converter to move over 4 VMs from VMWare Server and then carried out the performance tests.
We are committed VMWare fans, so based on the feedback received here, we'll go back and try again. This time we'll try new VMs created directly on ESX and see if maybe the conversion process caused the problems.
sounds like a good plan, i have had problems with P2V stuff before (i know u were doing v2v), you should get much better performance on on ESX.
do u know where the contention is occuring? can you run an esxtop and post the results back here, we may be able to pin point the problem