I have two servers. One DELL R510 with PERC6/i (megaraid_sas driver), and one DELL R210 with PERC H200 (mptsas driver). They're both configured as RAID1 (mirroring).
On the R510, guests have great performance. ESXi reports a datastore write latency between 0 and 2ms. Wonderful.
However, on the R210, guests are incredibly slow, due to I/O waiting. "top" shows 20% of "wa" (I/O wait), and guest load is usually around 0.20 (while doing nothing!). Maximum write speed is around 6 Mb / s. Not really usable.
ESXi reports datastore write latency between 200ms and 500ms. That explains bad guests performance. Is this a known issue with this hardware ?
RAID status is "OPTIMAL", so the latency does not come from a recovery.
What can I do ? Are there some hidden configurations or actions that could be done via ssh or through the vSphere Client to get decent performance ?
Thanks for your help.
Message was edited by: florent42 - added OPTIMAL state
Welcome to the Community,
from what I read about the RAID controllers you mention, The PERC 6/i has 256 MB BBU cache and the H200 does not have cache (at least no write cache). Therefore it operates in write-through mode what most likely causes the latency you see.
The only solution I see is getting another RAID controller with BBU which can operate in write-back mode.
EDIT: Link to the DELL RAID controlles
On a PERC H200 card, caching is forced to be disabled for all physical disks configured into a virtual disk, regardless of the drive type and default drive settings.
I understand there is RAID controller cache (not present on H200) + physical disk cache (disabled on H200), which explains the bad performance.
Thank you, I will look into a PERC 6/i controller.
The H200 controller disables all disk caching built into the drives by default.
If the server has battery backup you can enable basic drive caching on the Logical Volumes without risk.
Goto www.LSI.com (The controller is made by LSI) and download and install MegaRAID Storage Manager.
Highlight the Virtual Raid Volume, right click, properties. Enable Cache and save.
You have to do this on every logical raid volume.
Your performance difference is incredible.