In my earlier blog about choosing a storage controller, I mentioned that the Microserver's on-board AMD SB820M SATA controller doesn't quite let the SSDs perform at their best.
Just how bad is it?
I did run some tests with the fio benchmarking utility.
Lets have a look at those random writes, they simulate the workload of synchronous NFS write operations:
rand-write: (groupid=3, jobs=1): err= 0: pid=1979 write: io=1024.0MB, bw=22621KB/s, iops=5655 , runt= 46355msec
Now compare it to the HP Z800 on my desk, it has the Crucial CT512MX100SSD1 on a built-in LSI SAS 1068E controller:
rand-write: (groupid=3, jobs=1): err= 0: pid=21103 write: io=1024.0MB, bw=81002KB/s, iops=20250 , runt= 12945msec
and then there is the Thinkpad with OCZ-NOCTI mSATA SSD:
rand-write: (groupid=3, jobs=1): err= 0: pid=30185 write: io=1024.0MB, bw=106088KB/s, iops=26522 , runt= 9884msec
That's right, the HP workstation is four times faster than the Microserver, but the Thinkpad whips both of them.
I don't know how much I can expect of the PCI bus in the Microserver but I suspect that any storage controller will help me get some gain here.