[[PageOutline]]
= Lustre Performance Report (Uses iozone command) =
[[BR]]
== 0. Machine Informance ==
||Node ||4 nodes (1 server , 3 client provide disks)||
||CPU ||Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz (each node)||
||Memory ||2GB DDR2 667 (each node)||
||Disk  ||320G+160G (each node), All nodes:  (320G+ 160G) * 3 = 1.4T||
||NIC ||Intel Corporation 82566DM Gigabit Network Connection||
||Switch ||ZyXEL GE switch||

== 1. 4 Nodes, No Replicate, Adjust Parameters ==
{{{
$ iozone -g 16g -aRb test.wks
}}}

[[BR]]
== 3. Performance ==

The following six diagrams show the testing results using IOzone as the benchmark to evaluate how Lustre works with and without data-replication. The six IO operations we have done for performance evaluation are write, re-write, read, re-read, random read, and random write opertion, respectively.

[[Image(write.PNG)]]

[[Image(re-write.PNG)]]

[[Image(reader.PNG)]]

[[Image(re-reader.PNG)]]

[[Image(random-reader.PNG)]]

[[Image(random-write.PNG)]]

[[BR]]

 * 原始檔案請下載附加檔案 --> lustre_iozone_16GB_single.xls

 * 極值分析
|| Report || Record Size (KB) || File Size (KB) || KB/s ||
|| Writer || 64 || 32768 || 985975 ||
|| Re-writer || 1024 || 1024 || 1450281 ||
|| Random Write || 256 || 32768 || 1342730 ||
|| Record Re-write || 512 || 8388608 || 2382095 ||
|| Fwrite || 256 || 16384 || 1052347 ||
|| Re-fwrite || 256 || 512 || 1656957 ||
|| Reader || 1024 || 1048576 || 2686370 ||
|| Re-reader || 1024 || 1048576 || 2780881 ||
|| Random Read || 1024 || 1048576 || 2699113 ||
|| Backward Read || 1024 || 1048576 || 2749793 ||
|| Stride Read || 1024 || 1048576 || 2747632 ||
|| Fread || 512 || 1048576 || 2664701 ||
|| Re-fread || 1024 || 1048576 || 2701819 ||