Recently I have done some informal disk performance test on my newly arrived workstations. (copy a 1GB text file on the same file system).
On a Dell Precsion T3500 with Linux software RAID1 of 2x500GB SATA disk with AHCI enabled :
[root@gsx42 disk1]# time cp a b
real 0m4.499s
user 0m0.054s
sys 0m1.367s
On our busy mail server with PERC6i+ 2x15krpm SAS disk
[root@server mail]# time cp a b
real 0m57.504s
user 0m0.209s
sys 0m2.859s
It seems that the R/W performance of our PERC6i+SAS disk is not very satisfactory when compared to our new workstation though the mail server is much busier. After some research on google, I added the below parameter : blockdev –setra 16384 /dev/drive. The original readahead buffer for Linux is 128k, I changed this to 8M and redo the test :
root@server mail]# time cp a b
real 0m36.426s
user 0m0.398s
sys 0m2.673s
A nearly double in performance!! I also note that this readahead buffer is only useful in sequential read of large files. This may not be useful for lots of random access of small blocks of data. However, for our mail server, it need to handle lot of large mailboxes so this should be useful. And for Vmware server hosting the vms in local storage, this parameter will be useful too.
Useful references :
1. http://webcms.ba.infn.it/cms-software/cms-grid/index.php/Main/TuningDiskServers
2. http://www.overclock.net/linux-unix/388475-linux-performance-tuning.html
Leave a Reply