At 12:15 11-2-2001 -0600, Steve Lord wrote:
We are not having a good day here in XFS land!
Neither am I, tommorow morning I am positively sure that nox faxes will be
sent... ugh
> hi,
>
> I have a machine at work which is a PIII 450 with 128MB ram and a 15GB IDE
> disk that will oops on the moment That I am trying to write large amounts
> of data on a samba share.
> The samba share is located on /home which is a XFS partition most of the
> times it happens about the time that the system runs out of ram buffer (eg
> 100MB) and then oopses mentioning swapper.
Samba itself does not do anything special - it just does writes to push
data out. Are you using kio or kiocluster in the mount options?
Nope, straight mounting without any extra options.
> 2.4.0-XFS (before 2.4.1 merge) does boot and takes the punishment with
ease.
> Unfortunately I can not send the oops because:
> - The file is on the disk there.
> - The machine is currently displaying another oops
> - I can not reboot it on untill tommorow when I get to work.
>
> I can also trigger a oops using the Eicon Diva server which it contains.
> The drivver is lost after sending a fax. Open minicom on the isdn device.
> Modem init goes OK and as soon as I type atdt 000 and enter it oopses.
> Although the oops says swapper is involved I don't see the link.
> These are things that started happening after the 2.4.1 merge.
> This machined has been rock stable for the past testing period.
This problem has got to be a generic linux problem - not xfs (I hope).
I don't know for sure. I can replicate the Oops using the Eicon Diva server
ISDN adapter ($500) using 2.4.0
I am positively sure that something in the 2.4.0-2.4.1 merge went wrong,
horribly wrong.
> Sarcastic side Note: This $1800 desktop machine with samba beats a $10.000
> dell poweredge 2400 with twice the amount of ram and disk space with raid
> 10 on 4 18GB 10KRPM disks. ... by a factor of 5 that is using the
2.4.0-XFS
> kernel.
Can you elaborate on this - is this an XFS is slower than yyy or is this a
dell is slower than xxx?
It is more of a puny linux desktop beats the pants of a big NT server.
XFS speed is speedy by my standards. And the error recovery is even more
appreciated then the faster disk IO. It justs takes 2 minutes to come up
after a crash. That's worth a lot more to me.
The linux box is a Dell Optiplex Gx1 PIII 450 with a standard Maxtor 5400
RPM ide disk and a 3c905 network adapter connected to the same 3com
SuperStack 3300 that the Dell poweredge is connected and our IT department.
100Mbit Fast ethernet back to back over this switch.
The NT box is a Dell Poweredge 2400 with a PIII 733EB and 256MB ram using 4
10K rpm 18GB Ultra2 scsi LVD disk drives connected in raid 0+1 (raid 10) on
a Ami Megaraid controller. The system is equiped with a Intel Etherexpress
100+ server adapter.
Beaming over 2.8GB of binary data from a laptop (windows 2K) to the NT
server took 56 minutes. Beaming it over to the linux machine took just
under 10 minutes.
When beaming over the file from the linux box to the NT server it took
about 10+ Minutes
So it seems that the server part is relatively slow.
When using Norton ghost on this laptop to restore that disk image to a new
machine it would clock a 80MB/min restore rate when streaming the Image
from the image located on the network share of the NT machine.
When using the image located on the samba share of the linux box and
restoring we clocked a new restore speed record of 200+MB/min. We never
restored images residing on a linux box before, so this was a first. And
the old one was at 120MB/s.
Very cool indeed. This is not directly XFS related. It just shows that both
linux 2.4 and 2.2.18 are very fast in networking when coupled with samba.
These things bring back netware memories (decent speed)
And now the baby monitor is calling, I have to go be 'da da' ;-)
Cool, good luck :-)
Steve
--
Seth
Has anybody seen my lightbulb?
I _really_ need some light here.
|