I was given the task (aka dumped into my lap) to examine low Read/Write performance on a system running Server 2012 R2 /w HyperV role.
The system in question:
Lenovo RD340
LSI MegaRaid 9260-8i (no BBU)
-Drive Group 0 consisting of 2 SAS Drives, each 279.397 GB, RAID1
-Drive Group 1 consisting of 3 SAS drives, each 558.912 GB, RAID5
Virtual machines are stored on drive group 1.
VMs: Two domain controllers, a terminal server & an application server
It seems that read/write performance for drive group 1 is quite low. For example, when I try to export one of the VMs via HyperV Export onto the same drive (BYA: necessary for scripted exports), the max throughput is around 25~30 Megabytes/s. Seems low, especially in comparison to another system (Lenovo TD340 with LSI 9260-8i (with BBU)) where such an export nets around at least 150 MB/s and more.
Any idea what I should look out for to maximize performance?
Further settings for drive group 1:
- Strip size 64 KB
- Read Policy: No Read Ahead
- IO Policy: Direct IO
- Current write Policy: Write Through
- Default Write Policy: Writhe Through
- Current Access Policy: Read Write
- Default Access Policy: Read Write
The "control group" drive group on the RAID controller in the TD340:
- strip size 256 KB
- Read Policy: Always Read Ahead
- IO Policy: Direct IO
- Current Write Policy: Write Back
- Default Write Policy: Write Back with BBU
- Current Access Policy: Read Write
- Default Access Policy: Read Write
Thanks a lot!
Best Answer
From my measurements done a while ago on 9240 and 9260 LSI Megaraid cards in UCS servers the I/O perfomance using RAID 5 and 6 was significantly lower than in RAID 0 or 10, everything else being the same.
My suggestions would be (depending on the tolerance to failures in your context) either:
Using WriteBack instead of WriteThrough will boost write (and, depending on usage, even read) performance, regardless of the RAID type. It needs to be specifically enabled if you don't have a BBU or if it goes bad. A BBU is highly recommended :)
Finally I second @SteffenNielsen's comment about the write cache, also regardless of the RAID type.