Server/SQL 2000 ESXi 4.1 Strange Issues

[QuazzieM]

Hey guys,

I'm hoping that someone might be able to shed some light on the limitations/issues that I have found since virtualising one of production servers for TESTING purposes. Please excuse if I user the wronge terms, in my post... Farily new to the VMware scene..

To start, I have a ESXi 4.1.0 server running on a Dell R810 (dual Intel E7540, 128GB RAM) these are connected through10GB iSCSI NICs to Dell EqualLogic SAN.

The server I needed to get up and running is a old version (dont ask to why were using an extremely old version of this app.. Long and painful story) of JDE application server running on Server 2000 Enterprise, also running SQL server 2000.

I created the virtual server using the converter tool, which went perfect. Its up and running and has its SQL data volume attached and the JDE application is up and running. (Note the VM and its datastore are all located on the EqualLogic SAN)

Now my issue is that the JDE application has some internal jobs that are highly SQL intensive which it does overnight. On the curren live server it takes 5hrs to complete usually.

On our virtualised server it has some very strange and varied results which I believe has something to do with how its using the vCPU's for its multithreading.. (I think thats the correct term)

If I run the 5 hour job on the VM having different vCPU's settings I get the following results:-

1 vCPU - 6 hrs to complete the job

2 vCPU - 5.5 hrs to complete the job

3 - 8 vCPU - Job doesn't seem to complete (even after 24hrs) its just in an endless run. (I can see that there is IO and MB throughput on the EqualLogic, but not much at all)

Is there some advance settings that I should be tweaking to possible help fix this issue? or it simply just a known issue of virtualising server 2000?

Cheers

[golddiggie]

First thing I would say is that this is a prime example of why you want to keep vCPU count to a minimum on VM's... Tossing more vCPU's at a VM doesn't mean it will run better... Most of us run VM's with either one or two vCPU's allocated (in extreme cases, maybe four vCPU's)...

Did you remove any ghost hardware on the VM once the conversion completed?? Instructions to do that can be found in KB article 1179

Also, have you properly aligned (64K partition block alignment) the volumes on the VM?? That can impact the I/O of the VM... You could give VizionCore's vOptimizer Pro a try for that... Otherwise, you'll need to do it the hard way (couldn't locate the KB for that, believe there's at least a couple of threads here for how to do it)...

How much memory have you given the VM?? Are the resource shares the default values, or have they been changed?

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[QuazzieM]

As mentioned I had originally created the VM to use a single vCPU due to limitations found in other threads regarding verbalizing server 2000 However after running some tests I found that fobs were take 20% longer to complete and the CPU usage sat at 100% constant whilst the jobs were running. As this bottlenecked the server so no other operation could be done whilst they're running I thought that I would up the vCPU's to 2.

This resulted in a slightly faster processing of the job's but nonetheless it was still slower then on the current physical server (which is a few hardware generations older then the ESXi server). But when I upped the vCPU's to anything above 2 the jobs wouldnt complete.

I had removed all the old ghost hardware as per the KB article you mentioned and also made sure the disk alignment is correct. Which rules out those posibilities, also actual SQL queries seem to 30% faster on the VM then on the live, so its only these larger jobs (which use UBE to start) that seem to be taking a longer period to complete then on the live system.

The only other thing I can note is that when I'm using a single vCPU, in the Resource Allocation tab for the VM, its sits at 100% consumed at 2GHz, is there a way to give it more CPU power without enabling more vCPU's?

[golddiggie]

Using the conversion tool is not the same as creating a VM from scratch...

When you modified the VM to go from the original processor (or core) count down to one, did you modify the HAL to reflect it was now an uniprocessor system? Did you correct it when you went back to a multi-processor setup? That is a rather important item. Especially when you're talking about win2k server...

For S&G's you could build a fresh Win2k server VM, install the SQL version/edition you need to run, then attach the database to it to test out performance... I would make sure the disk of the new VM was aligned from the start (before installing the OS perform the alignment on the VM, then on the additional drives as you create them)...

I would also make sure you don't have any software lingering that was only relevant on the old hardware. Things like video card drivers/utilities, old network card software, etc... You'd be amazed at how much something that simple will bugger a VM (or mess it up just enough to make it annoying)...

Is it safe to say that you've installed the VMware Tools onto the VM as well? Which SP level is the server at?? That, alone, could be a factor in how things are running for you... Especially when you're dealing with a decade old OS...

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.