El lun, 08-07-2002 a las 17:11, Matthias Andree escribió:
> "Roger B.A. Klorese" <rogerkqueernet.org> writes:
>
> > Matthias Andree wrote:
> > > I have no experience with XFS or jfs. My experience is restricted to
> > > ext2fs, ext3fs, reiserfs.
> > >
> > > If you value your mail, there is only one choice: ext3fs. It's the
> > > only(*) file system that has ordered data writes, that prevent corrupt
> > > file contents over a crash. It has journalling, and you can see if data
> > > journalling makes it faster for you.
> >
> > Unfortunately, without an external journal, with no tuning, and on top
> > of a hardware RAID-5, ext3 proved too slow for us to run on -- kjournald
> > ran at 99% CPU.
>
> kjournald at 99% CPU? Hard to believe, never seen that, and all Linux
> boxen that I operate have at least some ext3 file systems, but none has
> RAID-5. Either there is a kernel bug or RAID-5 and ext3 don't work
> together well. May I ask what hardware RAID you're using?
>
> The Postfix server with the longest uptime that I look after has been
> running since 107 days 8.5 hours, and the most expensive kjournald
> process has taken 46 minutes of CPU time (Pentium-II/266 -- Klamath
> core), the next nicer 8.5 minutes, the next 5.5 minutes and the other
> two are well below a minute. That's less than 0.04 % CPU. AIC7xxx-based
> UWSCSI, Linux-2.4.19-pre1-ac2, only ext3 file systems.
>
> May I recommend to try the latest *.rc or *.aa kernel version and see if
> the problem persists? If it does, may I suggest to report that to the
> linux-kernel mailing list? Some people on that list may want know about
> your problem and may be able to help you. Please send them exact
> information on your hardware and the kernel version that you're running
> so they won't have to ask too much. Patches applied, etc. If you can
> provide a test case to trigger that 99% CPU condition in kjournald, the
> bug should be fixed pretty soon.
>

        A few weeks ago I was running in a trouble with a server with ext3,
when any process have to do some intensive IO to hard disk (for example
perform a mysqldump) and the load of machine start to increase more and
more and more (more than 100!!!!) and finally the server get in an
unknown status... :-S

        I was running RAID1 with two IDE (20GB and 40 GB, each disk) disks and
kernel 2.4.18. Then I started to play with renice, reniced to -20 the
kjournald process, and finally use hdparm to activate DMA and 32Bit IO
access (hdparm -c1d1 /dev....) and the server started working without
any problem .....

Certainly strange

Agustin

-
To unsubscribe, send mail to majordomopostfix.org with content
(not subject): unsubscribe postfix-users

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]