Feb 23 2011
MSA 2012i SAN Performance Windows vs Linux
I’ve noticed backups are getting slower and slower over time going from 500 to 2500MB/min throughput.
I run backups on 2 dedicated RAID0 disks hosted on a iSCSI MSA array first, and duplicate them on tapes. Since I want to use the same array for a new database project, I want to measure and tune access to SAN disks from Windows 2003 Server and Linux Redhat 5.
Environment
4 disks have been set up in a RAID 10 set.
All interfaces are forced in 1G speed on the SAN and the server side.
Servers are connected on to the SAN via 2 interfaces in a load-balanced modethat gives a max theorical throughput of 250MB/s (a bit less considering frame headers).
Jumboframes have been disabled, they’re not supported on the switches I’m using in this setup. They could give slightly better performance.
I also tuned the read ahead cache but I got no significant improvement.
Results
The tests were conducted with a simple dd on Linux and Diskbench on Windows.
Reads MB/s IO/s Windows 70 1100 Linux 110 900 Writes MB/s IO/s Windows 140 2100 Linux 140 400
Optimizations
There was no optimization whatsoever on the Linux platform. The partition was formatted with default ext3 filesystem on a LVM volume.
On the Windows side, there was no fragmentation since we started off with a brand new drive. Fragmentation degrades performance indeed. The disk is formatted in NTFS with default 4k clusters: Increasing the cluster size does not seem to be of much impact.
The partition has been aligned with the physical disk in diskpart to correct Windows 2003 Server caveat. This led a 5MB/s increase in the maximum disk throughput.
Then?
2 questions:
Why the HP MSA 2012i writes faster than reads!? Write cache?
Why Windows read operations maximum bandwidth gets stuck to 70M?
Throughput was roughly improved by 5M with above Windows optimizations but we seem to now reach a cap limit. I get the same results with whichever RAID0, RAID5 or RAID10!!