On Wed, Jul 30, 2003 at 08:47:16PM +0200, Seth Mos wrote:
> At 10:04 30-7-2003 -0700, Cliff Wells wrote:
> >On Wed, 2003-07-30 at 02:09, Seth Mos wrote:
> >> At 18:31 29-7-2003 -0700, Ravi Wijayaratne wrote:
> >> >Hi,
> >> >
> >> >We are seeing a %15 performance drop when we move from XFS 1.1 to 1.2.
> >> >Here are some of our particulars:
> >> >We have been testing the performance of Linux-2.4.19 and XFS 1.2 with
> >> >NetBench.
> >> >We compared the performance with that of Linux-2.4.18 and XFS 1.1.
> >> >We have been running Samba 3.0.
> >>
> >> The only problem I know of which is not exclusive to XFS (ext3 does
> >this as
> >> well) is one on the read performance of 2.4.18 vs the 2.4.20 errata
> >kernel.
> >>
> >> On 2.4.18-27 I get 170MB/sec read speed
> >> on 2.4.20-18 I get 80MB/sec read speed.
> >>
> >> the write speed stays consistent between these releases but the enormous
> >> drop in read speed is very strange. It appears to be present in the
> >> standard rh errata as well.
Sorry if this is a repeat, I haven't seen the whole
thread. Did you try increasing the scsi max readahead?
echo "511" > /proc/sys/vm/max-readahead
echo "127" > /proc/sys/vm/min-readahead
This helps on my filesystems with Fibre Channel hardware.
However, I have always had this problem. It didn't start
with 2.4.20. I don't use Redhat kernels so maybe there
is something that changed there.
Craig
> >
> >Try hdparm -a 512 /dev/hdx
> >
> >http://marc.theaimsgroup.com/?l=linux-kernel&m=105830842018800&w=2
>
> That's IDE. I am talking about scsi. The 3ware controller is just a scsi
> controller with IDE disks, but the OS doesn't know what disks are behind
> the raid container.
>
> [root@lsnetniet root]# hdparm /dev/sda
>
> /dev/sda:
> readonly = 0 (off)
> geometry = 45051/255/63, sectors = 723757824, start = 0
> [root@lsnetniet root]# hdparm -tT /dev/sda
>
> /dev/sda:
> Timing buffer-cache reads: 128 MB in 0.66 seconds =193.94 MB/sec
> Timing buffered disk reads: 64 MB in 1.06 seconds = 60.38 MB/sec
> [root@lsnetniet root]# hdparm -a 512 /dev/sda
>
> /dev/sda:
> setting fs readahead to 512
> BLKRASET failed: Invalid argument
> readahead = 120 (on)
> [root@lsnetniet root]# uname -a
> Linux lsnetniet.coltex.nl 2.4.20-18-rh-xfs #2 SMP ma jun 23 11:01:51 CEST
> 2003 i686 unknown
> [root@lsnetniet root]#
>
> And thus the experiment fails.
>
> Any other suggestions :-)
>
> Cheers
>
> --
> Seth
> It might just be your lucky day, if you only knew.
>
--
Craig Tierney (ctierney@xxxxxxxx)
|