How about trying 32bit linux host OS aswell?
Because I don't have any spare partition space to put a third O/S! But I agree it would be interesting to see how this compares too.
... deleted ...
(Just noticed I was answering to the wrong topic alltogether. Sorry 'bout that.)
Don't know, and it could be anything before you blame memory performance in the Linux host. I have SQL Server running on a Windows 2003 guest running on a CentOS 5.2 host and I haven't noticed anything untoward. To be honest, my impression of performance within guests running on a Windows host has not been favourable.
Could you try and narrow this further as a Linux problem by using another host OS, possibly Red Hat or CentOS?
...We used a benchmark tool in the guest called Sandra Lite2009...
According to the "Performance Best Practices and Benchmarking Guidelines" at ( http://www.vmware.com/support/pubs/ws_pubs.html ):
We don’t recommend use of the following benchmarks, as our experience has shown that they can produce
inconsistent results in virtual machines:
Granted the guide was written for VMware Workstation 6.5, but I'm sure it also applies to VMware Server.
Thanks for that link Scissor. Even though Sandra might product inconsistent performance running in a guest, we were doing a comparison of the same guest running in different hosts so I would have thought there would have still been some relativity about the results, but we'll take the advice and try one of the recommended ones.
Will post back results ...
Using the lite (free) version of PassMark's PerformanceTest (one of the packages recommended in the document), gave much more consistent results. Clearly the timing measurements in Sandra must be screwy when used in a guest VM.
This was only a real quick and dirty comparison, but:
For the memory speed tests, we found the Windows guest running under Suse 11 (64bit) was about 5 to 10% down compared with Win2003 (32bit) as the server.
CPU performance was pretty much line ball between the two.
Disk performance ...ntfs performance (for the vdisk) under linux was down about 20 to 25% on windows host. But using ext3 - as you would do - puts the two contenders on a par, but results are quite dependent on the block size used in the tests.(The bigger the block size in a sequential write tes, the better ext3 compared to ntfs). Somewhat surprisingly to me, these tests became progressively cpu bound with larger block sizes.
Also compared these results with the same tool running in the (Windows) host itself, and the guest performance overall was only degraded by a small and acceptable margin compared to the host.
So all in all I'm not unhappy with these results, although somewhat puzzled why host=linux-64 is still slightly lagging performance of host=windows-32, when I had expected it to be the other way around based on what others have said.
I would be very interested in these sort of comparisons from anyone else...just to see if these results are typical.
I guess you mean this thread!
"The big-picture answer is that the Linux virtual memory subsystem is simply not tuned for running VMs. No OS can be the best at everything. If you really want to be at the higher ends of transfer rates for your storage devices, ESX is a much better option - ..."
After this I was thinking to test virtual box on my ubuntu host - but I really like vmware server and with only a few vms the performance is ok for me though the windows vms are quite slow....
Ah, yes, that is the one.
I haven't tried this out with a whole lot of Windows VM's, but at the moment the two host O/S choices seem to offer close enough performance that I will be happy to run with Linux. I will also probably use a number of Linux guests for things that currently are Windows, simply to make it leaner and meaner, also to take advantage of paravirtualisation if I can.
I am running one ubuntu 8.04.1 jeos guest with para virt support. it performs really great... and I just love debian based distris and with 5 years support its my choice for linux guests....
I'm interested to hear you say that...I have actually built a guest VM with JeOS when 8.04 was first released, and I really like that idea.
Do you have any sort of comparison of performance for something 'real worldish' such as a J2EE server, or just Tomcat/Servlets/JSP/Java between JeOs and Windows on the same host?
I started out looking at using Ubuntu for host, but ended up going down the SuSe road for a couple of reasons. But Suse don't seem to have an equivalent skinny build of the O/S to match what JeOS aims to do. I'm assuming it shouldn't matter that I might use JeOs as a guest on a SuSe host? Also, why is there no 64bit JeOS build?
Sorry, cant offer any real world performance comparison - its running an ssl gateway - java based. Compared to the windows guest it really "feels" better - just like a physical machine....
My host is running ubuntu 64bit - but I think you are right - it doesn't matter what linux distri you are using on your host.
And I have no clue why there is no 64 bit version of jeos...