Busy disk and page fault
Hi all, I'm asking myself about a problem I have with a Postgresql server on FreeBSD 7.1-RELEASE. The server is overload, the disk is 100% busy with 250 write operations per second and a throuput of 6MB/s. My first idea is because of mass random access/write on the disk. But I also see the server can make 20k page fault per second. So, did you think I really have a disk contention or this high number of page fault can be a problem (and if it can, how to resolve it). Thanks, -- Nicolas Haller ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"
Re: Busy disk and page fault
The server is overload, the disk is 100% busy with 250 write operations per second and a throuput of 6MB/s. My first idea is because of mass random access/write on the disk. But I also see the server can make 20k page fault per second. what page fault? most page faults in FreeBSD doesn't mean disk access, just no mapping present in page tables, which gets mapped after the fault. Only if page is actually not present in memory it is fetched from disk. top shows in what state is a process. if it's biord or biorw - it's doing disk/file I/O, not swapping. that's about FreeBSD part - about postgress part ask on postgress mailing list. i don't use it so i can't help you. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"
Re: Busy disk and page fault
On Mon, Mar 09, 2009 at 05:04:16PM +0100, Wojciech Puchar wrote: >> The server is overload, the disk is 100% busy with 250 write operations >> per second and a throuput of 6MB/s. >> My first idea is because of mass random access/write on the disk. But I also >> see >> the server can make 20k page fault per second. > what page fault? most page faults in FreeBSD doesn't mean disk access, > just no mapping present in page tables, which gets mapped after the > fault. > Only if page is actually not present in memory it is fetched from disk. > top shows in what state is a process. > if it's biord or biorw - it's doing disk/file I/O, not swapping. The box don't swap. I just ask if page fault interrupt postgresql process and fragment/de-optimize disk write. -- Nicolas Haller ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to "freebsd-questions-unsubscr...@freebsd.org"