i/o bandwidth on MSA2212fc attached to sles9 blades

[joseph pareti]

I need to demonstrate that an infrastructure consisting of n c-blades attached to a SAN is capable of delivering > 40 MB/sec sustained write bandwidth to users (5 users on each of the 8 blades). I also need to interpret the results, see my question at the end of this post.

The benchmark is based on making a copy from an internal disk to the SAN. Because the real workload calls for file systems usage we cannot use raw device copies.

The configuration is as follows:

- 8 c-BL460/SLES9,
- 2 emulex MultiPath Ports per blade connected to the switches
- 1 MSA 2212fc:
4 FC links from the MSA to the switch
48 disks
- 4 virtual disks, ext2 file systems on each of those
Volumes on pl241,2,3,4:
/ext2_test1 is a volume on vdisk1
/ext2_test2 is a volume on vdisk2
Volumes on pl245,6,7,8:
/ext2_test1 is a volume on vdisk3
/ext2_test2 is a volume on vdisk4

Per MSA2212fc specs RAID10 is capable of 474MB/sec sequential write. This would translate in 474/45~10 processes (=users), each having 45 MB/sec
The test on 4 servers (simulating 20 users) achieves in average > 40 MB/sec; buffer cache effects might have helped?

[Steven E. Protter]

Shalom,

Buffer cache figures will have varying benefit based on what kind of applications you are using.

An Oracle db will benefit from more SGA and smaller buffer cache. mysql might see the opposite benefit it.

SEP