Currently I am running Zentyal Server(Ubuntu) as Guest on a esxi 5 host on a Adaptec raid controller RAID 10 with Enterprise WD hard drives, 8 Core CPU and 8 Gb of ram (I know a bit overkill) to server about 40 Users. Currently 300GB of storage is being used out of 800GB on Thin Provision.

At the start Read/Write Performance was amazing (200MB/s read, 170MB/s write) but as the storage grow performance has come down to 9MB/s read and about 2-3MB/s write using dd and hdparm.

Cannot understand what could cause this, Any help would be appreciated. I have checked CPU and not much of it is being used unless someone is writing to it(maxes out, no antivirus on the server).

This seems to only happen on this Guest as I tried to create another Guest and hdparm and dd returned average 200MBs/170MBs read/write performance.

  • Can you give the adaptec config? I use 5805, BBU, write through and easily get 60mb on a 8 disc raid. The new 7805Q could use an SSD for caching. But one will need to get more statistics (which adaptec does provide) to help with the analysis. – TomTom Feb 15 '13 at 13:32
  • Hi Tom, It's Adaptec RAID 2405. I am using it with other servers too and haven't seen any problem. Just this Guest that I have problems with. Not sure if it is because of large storage. – Reza Feb 15 '13 at 13:33
  • Sorry, that is not large storage. The 2405 is supposed to run 100 discs or so. Check stats, and cabling. This really is too slow, unless something hits you totally crazy. – TomTom Feb 15 '13 at 13:35
  • 4
    Don't mind TomTom, he's just used to running websites getting 4 million hits per second on his customized iPhone, which has 14 zottabytes of storage. – HopelessN00b Feb 15 '13 at 14:03
  • Haha, As long as I get my answer! – Reza Feb 15 '13 at 16:10

I have found the solution to this problem:

  • Installing VMware Tools on the Ubuntu guest increased the performance a little.
  • The performance was significantly reduced due to the fact that I had assigned the Guest 4 virtual Sockets (vCPUs), following some research I found out that this causes big problems with Ubuntu (Possibly Linux) operating systems and causes longs switches and delays in the system responsiveness. After taking down the Virtual Sockets to 1 the OS performances hit it's peak performance instantly after a restart.

New Readings:

  • Using hdparm: -- Average Cached reads: 13255 MB/sec -- Average Buffered disk reads: 146 MB/sec

  • Using dd: -- Average write speed(bs=16k, count=256k): 169MB/sec

I do know that write speed shouldn't be higher than read speed but as we all know dd is not as accurate as it should be! But still a big increase from the pathetic average speed of 9MB/sec read and 3 MB/sec write.

Hope this helps those who have this issue.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.