Problem with GFS read and write performance
 

Hi,

I am using iozone to obtain read and write performance of GFS. I have a single logical volume with the GFS filesystem and mounted it on one server. I have used lock_nolock locking protocol to ensure it is treated as local filesystem.
When i run multiple threads of iozone on the GFS partition, I observe that the performance(read and write) numbers decreases as the no.of iozone thread increases.
Is this expected behaviour ? If yes, then how do check for scalability because as we increase the iozone threads the performance numbers would be decrementing.

Thanks in Advance,
RBH

Hi,

Do all the connecting nodes have the same date/time? If not (a difference of mere minutes is enough) there are a lot of unnecessary i-node updates.

Not much to go on, but hope this helps.