Subject: Performance Issue under RedHat

[Emil Henry]

Hello!

We have our storage LUN's coming from an EVA4400 and presented over fiber to DL380's running RHEL 5.6. The LUN is about 450GB of student files, and formatted to GFS. We were noticing that the HP Data Protector was taking a really long time to do the initial scan and the backup. It took about 12 hours to do the full backup, when it use to take about 4 hours under HP-UX. Therefore, we compared the difference between EXT3 and GFS on the RedHat machines, and it was significant.

Therefore, we setup a 2-node cluster running RHEL 6.2 and formatted the 2 test LUN’s to EXT4 and GFS2. We did a simple “time  tar -cvf /dev/null “ of each of the filesystems.

EXT4: 28 minutes

GFS2: 305 minutes.

RedHat Support had us try a “dd of=/dev/null if=/dev/mapper/mpathc bs=4096” on both filesystems.

GFS2: 45 minutes

GFS2 with lock_nolock: 41 minutes

EXT4: 38 minutes

Therefore, I am assuming that the issue is with RedHat, and am looking for thoughts. RedHat Support has not been very good in getting this resolved.

Thanks in advance.

Regards,

Emil

[Dennis Handly]

>RedHat Support had us try a “dd of=/dev/null if=/dev/mapper/mpathc bs=4096”

Are you sure it wasn't /dev/zero?

[Emil Henry]

Hi Dennis!

I have used only /dev/null as the output file. Never tried /dev/zero. We were trying to read existing data from the file system and writing to a null device, instead of writing to the tape library.

Emil

[Dennis Handly]

>I have used only /dev/null as the output file.

Oops, sorry, I assumed the input (if=) was first.