xfs
[Top] [All Lists]

Re: Fragmentation of Journaling FS

To: Utz Lehmann <leh@xxxxxxxxxxxxxx>
Subject: Re: Fragmentation of Journaling FS
From: Constantin Loizides <Constantin.Loizides@xxxxxx>
Date: Tue, 07 Aug 2001 12:49:07 +0200
Cc: xfs-list <linux-xfs@xxxxxxxxxxx>
Organization: Innovative Software AG
References: <3B67D35E.64877CBF@isg.de> <20010801144322.C1132@de.tecosim.com>
Sender: owner-linux-xfs@xxxxxxxxxxx
Hi Utz,

>[..]
> I made a quick and dirty test running this:
> 
> while time cp -a /usr/src/linux/drivers/ /mnt/xxx-`date '+%s'`; do sync; \
> df | grep mnt; done
> 
I tried your crude test on my machine with the 512 MB partition. It also
shows
smooth behaviour. That made me a bit nervous, so I played around with
the configuration of the directory structure and gave agesystem another
run:

You find my results on the web
http://www.informatik.uni-frankfurt.de/~loizides/reiserfs/agesystem.html#gaugeveri

It seems that the sharp performance drop is influenced by the directory
hierachy
I am using. Why is the best behaviour for the least directories (100x1
means 100 dirs)?


I am also reproducing for the 4GB partition, but it will take some time.
The 100x50 test is alread there, compare it with the 100x100 test you
find at
http://www.informatik.uni-frankfurt.de/~loizides/reiserfs/agesystem.html#4part  

The results seem also be highly dependent on the directory structure.
Will also
test 100x10 and maybe 100x1 too.



Constantin


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