xfs
[Top] [All Lists]

Re: nfs/local performance with software raid5 and xfs and SMP

To: Tru Huynh <tru@xxxxxxxxxx>
Subject: Re: nfs/local performance with software raid5 and xfs and SMP
From: Steve Lord <lord@xxxxxxx>
Date: Fri, 20 Jul 2001 11:00:36 -0500
Cc: Nathan Straz <nstraz@xxxxxxx>, "linux-xfs@xxxxxxxxxxx" <linux-xfs@xxxxxxxxxxx>
In-reply-to: Message from Tru Huynh <tru@xxxxxxxxxx> of "Fri, 20 Jul 2001 17:45:43 +0200." <3B585227.2715F310@xxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
> Hi
> 
> On the same "benchmark" theads of Nate Straz ...
> (sotfware raid5 and xfs performance issue)
> 
> How long does it take to extract the 2.4.6 linux kernel and delete it?
> 
>     | create   | delete
>     -------------------
> a)  |   0:17   |   0:07
> b)  |   0:24   |   0:25
> c)  |   1:51'  |  16:45
> d)  |   2:54   |   8:08
> 
> [tru@adm ~/tmp]$ du -sk /tmp/linux-2.4.6.tar.gz ./linux/
> 26312   /tmp/linux-2.4.6.tar.gz
> 130412  ./linux
> 
> a) local xfs partition on the will become nfs client
> b) local xfs raid5 array on a single disk (/dev/md1) on the will become
> nfs server
> c) local xfs raid5 array on 3ware card 7disks (/dev/md0) on the will
> become nfs server
> d) nfs 
> 

Just one comment here, the configuration in case c) is never a good idea,
the raid code will automatically cause the disk head to thrash around
that is why the performance is so bad there. I am not totally sure why
we can do a reasonable job on deletes on a single disk, but not on a
raid - there are synchronous writes in there. Unless your single disk
xfs is on ide with write caching turned on. One more comment on xfs
delete speed is that multiple parallel threads doing deletes will
run faster than a single thread doing deletes, so lots of different
threads doing deletes will not be as bad as a single large directory
delete.

Steve





<Prev in Thread] Current Thread [Next in Thread>