Re: Poor performance from file system
From: Kamus of Kadizhar (yan_at_NsOeSiPnAeMr.com)
Date: 12/18/04
- Next message: Frank Sweetser: "Re: What good is RAID1 if both disks fail?"
- Previous message: larwe_at_larwe.com: "Re: Cypress-based USB-serial converters and Linux"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Fri, 17 Dec 2004 18:25:48 -0500
On Fri, 17 Dec 2004 06:35:37 -0800, Peter Saffrey wrote:
> Hi thre,
>
> I seem to be having some strange performance glitches since upgrading
> my kernel from 2.6.6 to 2.6.9. When I'm writing large amounts of data,
> for example: moving several GB of music files from another machine or
> writing decoded audio files, the performance of my machine is erratic.
> Half the time it's working fine, copying across several 4MB MP3 files
> a second. The other half, the machine is completely locked up and I
> can't even type. It's so unresponsive, it's difficult even to use top
> to work out what's consuming all the CPU, but I think it's kjournald.
> It's almost as if everytime I go over a certain threshold of data
> added to the disk it needs to do some slow and costly operation to
> work out what I've done. Could this be the case? What can I do about
> this?
>
Interesting. I was about to post my Q when I read yours. My details:
generic 2.6.9 kernel, two soft-raid arrays. One consists of two IDE hard
drives, each on its own controller and cable, running in a RAID-1 array.
The other array consists of 4 SCSI drives, running SE in a RAID-5. Both
arrays are formatted in large (approx. 100 GB) partitions using ext3, and
both are almost full - nearly 90%.
I copied about 90 GB of data from the IDE array to the SCSI array, and
found exactly the performance issues you did. The machine would "lock up"
for a minute or two at a time, no response at all. top showed 60 - 90% of
cpu time in iowait. Overall throughput was relatively slow - the
operation took about 2 hours.
Interestingly, now I am copying the same data from the SCSI array to the
IDE array, and have no such problems. There are slight pauses - a few
seconds at most, but in general the performance is acceptable. top shows
30 - 70% in iowait. Also, I got a small number of SCSI bus errors on
writes, but none so far on reads.
On top of that, it looks as though the SCSI to IDE transfer will be about
twice as fast as the IDE to SCSI transfer was.... I would have thought
the opposite.
So I'm going to guess that something is broken either in the IDE or SCSI
drivers or in the RAID drivers or in the ext3 journaling system.
Any thoughts?
Cross-posting to comp.os.linux.hardware added.
--Kamus
--
o |
o__ >[] | A roadie who doesn't ride a mountain bike has no soul.
,>/'_ /\ | But then a mountain biker who doesn't ride a road bike has no legs...
(_)\(_) \ \ | -Doug Taylor, alt.mountain-bike
- Next message: Frank Sweetser: "Re: What good is RAID1 if both disks fail?"
- Previous message: larwe_at_larwe.com: "Re: Cypress-based USB-serial converters and Linux"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
