alexander_foti1
Contributor
Contributor

Odd I/O and CPU performance on client

Jump to solution

Hi all

 

I have been having odd performance issues on an ESX 6.0 host.

In particular with one VM. 

The host is a Dell R515 with PERC H700 raid controller. 12 6gb/s SAS 3.5" disks in a Raid 10 and 2 identical disks in a Raid 1

6 core Opteron 4180 CPU X 2 and 64GB of RAM

Each Raid has its own Datastore on it. 

The Virtual machine has one 18TB Thick provisioned Virtual disk (Raid 50 Dstore) and one 1 TB Thin provisioned virtual disk (Raid 1 Dstore).

The Raw IO performance testing on these disks is pretty good:

Raid 1

Raid 1Raid 1

Raid 50

Raid 60Raid 60

The issue comes when you put lots of IO on any of them. The VM performance slows to a crawl. (Server 2012 R2). Its system process takes up half the available CPU% and local ICMP Replies from another machine on the Lan increase from <1ms to 100-800ms:

icmp.JPG

I don't feel that the ESXTOP outputs for storage and CPU neccearily represent a problem, although I feel %VMwait is a tad high (although 16x VCPUS)disk.JPG

cpu.JPG

Here is a screenshot of Task manager when a backup is running (high io to both Datastores/Raid's)

tskmgr.JPG

 

Any input would be much appreciated. 

 

 

 

Labels (4)
0 Kudos
1 Solution

Accepted Solutions
alexander_foti1
Contributor
Contributor

Don't all reply at once .... 🙂

 

This is now sorted. After doing a bit of digging, it looks like the primary OS drive was thin provisioned, and had an outdated snapshot on it. 

Once the disk was inflated and the snapshot removed, the performance improved dramatically.

View solution in original post

0 Kudos
2 Replies
alexander_foti1
Contributor
Contributor

Further information, Additional tests and shown that only IO to the Raid 1 array causes the VM slowdown.....

0 Kudos
alexander_foti1
Contributor
Contributor

Don't all reply at once .... 🙂

 

This is now sorted. After doing a bit of digging, it looks like the primary OS drive was thin provisioned, and had an outdated snapshot on it. 

Once the disk was inflated and the snapshot removed, the performance improved dramatically.

View solution in original post

0 Kudos