Performance problems on ESX 4.1.0 with DL385G6

[DYNAVICS]

Hello together,

I've got a quite weird problem:

If I copy big files (15GB) directly inside the datastore at the local disk array, the performance is quite good (around 250MBit/sec). As soon as I try to copy a big file (similar size) within the guest OS (Windows 2003 Standard, 32Bit, 4GB RAM, with Buslogic and also ParaVirtual Controller), I do not get more than 40MBit/sec (combined read+write).

Configuration: DL385G6 with 12GB RAM, P410 Smart Array 512 RAM BBWC, RAID5 with 4x300GB 2,5" disks.

There are only two guests running on this OS with similar config.

The customer is not very happy with the situation as the old (very old, 5 and 10 years) servers provided a better performance than the virtualized ones.

Please help, all suggestions are very welcome.

[Scissor]

Hi and welcome to the forums!

Hello together,

I've got a quite weird problem:

If I copy big files (15GB) directly inside the datastore at the local disk array, the performance is quite good (around 250MBit/sec). As soon as I try to copy a big file (similar size) within the guest OS (Windows 2003 Standard, 32Bit, 4GB RAM, with Buslogic and also ParaVirtual Controller), I do not get more than 40MBit/sec (combined read+write).

Converting your 250 MBit/sec number into MB/sec yields 31.25 MB/sec.

Just to double-check your timing calculations, are you sure that you are copying a true 15 GB file and not a thin-provisioned 15 GB VMDK that actually takes up less space on the disk?

For example, below I have a "template" VM with a 20 GB thin-provisioned disk. running "ls -lh" shows that the .vmdk file is 20 GB, yet if I run "du -h" it shows the true 4.6 GB of data actually in use.

/vmfs/volumes/4a1f371d-b60e2320-8923-001b212b36c2/Win2003-Template # ls -lh *.vmdk
-rw-------    1 root     root        20.0G Jul 14 04:17 Win2003-Template-flat.vmdk
-rw-------    1 root     root          529 Jun 13 07:09 Win2003-Template.vmdk

/vmfs/volumes/4a1f371d-b60e2320-8923-001b212b36c2/Win2003-Template # du -h
4.6G    .

Inside your Guest you report 40 MBit/sec = 5 MB/sec.

Can you confirm that your Host's RAID Controller is configured to cache disk writes and that its Battery is charged? If the RAID controller isn't caching disk writes that will definately cause the slow performance that you notice.

The Buslogic controller has some performance issues in Guests, so I don't use it myself: http://kb.vmware.com/kb/1890 .

The paravirtual controller should give you good performance, but just for fun can you try reproducing your tests with the LSI Logic controller?

[DYNAVICS]

Hi and thanks for your answer,

Yes I am sure that the file is not only provisioned as it shows this when browsing the datastore where both values are being displayed.

The measures are taken from the performance indicator of the host shown in the vSphere Client.

The measures for the guest are corresponding between host and guest statistics.

There it is showing the values in KBit/sec

The values for HDTach v 2.70 (just testing read speed as this is the most important issue we have)are:

XP machine with very average configuration (around 300 Euro):

14,7ms access, 81.2MB/s max, 37,5MB/s min, 62,9 MB/s average.

Guest machine on DL385:

4,5ms access, 517,6MB/s max, 3,5MB/s min, 133,4 MB/s average.

This is even more weird as the low values are between 0GB and 135GB drive position and highest from there to 260GB.

Could it be that the problem is the disks are too big?.

Hmm, there is a second disk on this guest which has the same issue, low speed to the middle and high speed at the end of the test,