[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] GFS: more simple performance numbers



Single node performance is heavily dependent on the lock manager(!). How about with nolock?

I'm also see what could be described as abysmal performance with the GFS, even with nolock on a single system. I'm using the 20040913 snapshot that was packaged for Debian by Bastian Blank. For sake of comparison I was attempting to copy the linux-2.6.8.1.tar file to the GFS partition to attempt to reproduce this timing test, and it took nearly 5 minutes just to copy the one file. This is on some slightly older hardware (P3-800) but I can't imagine that GFS would require a supercomputer to run...maybe it does?


Raw I/O and other filesystems on the same storage device work fine, so I have to point the finger at GFS...but I don't know where to begin in trying to track down the problem. I guess I should try the latest code from CVS first, but I'd love to have any hints or pointers the list could provide...

Thanks in advance,

Matt Mitchell


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]