high avg disk queue length

[danmo82]

Hi..

i have just installed a WMware ESXI 4.0 on my IBM M2 (7947-ZGA) server.

on this machine i have 2 guest one running server 2008 SBS and the other server 2008 STD. x64 with SQL 2008 installed

i have a problem with my SQL performance on the 2008 X64 STD. it is running a small SQL database (4Gb) for a Economy program.

when i run a READ/WRITE tests in the economy program it is slow, compared to the server that was running the database before

if look at the performance monitor on the SQL im expering high avg disk queue length.

the server is installed with 16Gb of ram, 8Gbg has been committed to each guest.

the server is installed with 2 physical CPU`s 4 cores are committed to each guest

the server is installed with 8 SAS disks

2 disk (RAID1) system disk (both guest are installed on those)

3 disk (RAID5) data share, Exchange DB on SBS server, and SQL database for SQL server.

3 disk (RAID5) data share, Exchange LOG on SBS server, and SQL log for SQL server.

there is only 30-35 users on this net, so exchange DB and SQL DB on the same disks should not be a problem i think.

enyone who has a idea to solve the avg disk queue length ??

[J1mbo]

There are lots of issues here, but firstly for the queue length - does the server have battery-backed write-cache on it's RAID controller?

Next RAM - some needs to be left for the host. for a 16GB machine, look to allocate 6GB to each of these two guests max.

Then CPU - use vSMP only if the sustained processing demands of the guests warants this. I assume the box is dual quad-core - even so, I would allocate 2 vCPU to each guest only.

Finally the disks. Seperating the spindles in this way is generally not the optimum configuration; the goal for virtualisation is to maximise the overall throughput in IOPS terms, rather than ring-fence drives for specific tasks. Currently those RAID-5 volumes are a real problem; with only 3 spindles any write IO will need all three disks to service causing high write latency.

Say these are 10k SAS drives, for a 70:30 read:write workload this config will probably do something like 320 IOPS (2 volumes, 640 IOPS but not shared). If you re-configured the machine as an 8 drive RAID-10, the overall throughput would be increased to something like 1350 IOPS.

HTH

Please award points to any useful answer.

[danmo82]

hi.

sorry for the slow reply, have been out of the office this week.

ill try chenge the things u said.

think ill leave the RAID 1 as it is, and then change the 2 RAID 5 arrays to a RAID 10 array

[AWo]

Another issue is the partition alignment. Check with your storage vendor what he suggests.

W2K8 should be fine with the partition alignment if the storage uses block sizes of 4096 or a multiple of it.

But depending on the storage you use the VMFS partition alignment may be wrong (if you use EMC or NetAPP that should not be the case as vSphere makes the correct alignment here).

A wrong alingment of the VMFS partition start or the guest partition start leads to an increased I/O traffic filling up your queues.

AWo

VCP 3 & 4

Author @ vmwire.net

\[:o]===\[o:]

=Would you like to have this posting as a ringtone on your cell phone?=

=Send "Posting" to 911 for only $999999,99!=

Edited by AWo

[MattG]

Is this the same physical disk setup that you used for the physical versions of these installs?

If you find this information useful, please award points for "correct" or "helpful".

[danmo82]

MattG: im not sure what you mean by the "physical versions of these installs", the old SQL server that were a physical server has been decommissioned,