From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rogier Wolff Subject: Fwd: Re: Reiser FS performance thingy. Date: Thu, 7 Aug 2003 20:58:54 +0200 Message-ID: <20030807205854.A20124@bitwizard.nl> Mime-Version: 1.0 Return-path: list-help: list-unsubscribe: list-post: Errors-To: flx@namesys.com Content-Disposition: inline List-Id: Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: reiser@namesys.com, green@namesys.com, mfedyk@matchmail.com, reiserfs-list@namesys.com, copy@harddisk-recovery.nl > OK Rogier, please search through the archive, and post the URL of your > previous report. That will be most helpful. I searched my own mail archive, remembering that I got an Email from someone named Hans. Do you know him? :-) Roger. ----- Forwarded message from Hans Reiser ----- From: Hans Reiser To: Rogier Wolff Date: Wed, 07 Aug 2002 00:03:00 +0400 Subject: Re: Reiser FS performance thingy. X-Spam-Status: No, hits=-100.0 required=5.0 tests=A_FROM_IN_AUTO_WLIST version=2.01 Rogier Wolff wrote: >Hi Hans, > >We've got a big filesystem on some pretty fast disks running reiserfs: > >Filesystem Size Used Avail Use% Mounted on >/dev/md0 603G 466G 137G 78% /recover5 > >/dev/md0 on /recover5 type reiserfs (rw) > >Just a couple of minutes ago, we happened to have only 60G left on >that filesystem, and suddenly performance went down the drain. Like >write 500k per second, while normally we're able to read (and write) >around 100Mb per second off the disks. > >We know that you're not supposed to use the last 10% of a disk because >of performance reasons, but this seems like an enormous penalty. > >My impression is that the "search for a free block" routine in >reiserfs somehow uses the wrong starting point. It is customary to >start searching where you found a free block last time and not say "at >the ideal point if you consider the directory entry". Then you might >end up searching 500G worth of block free flags for a spot for EVERY >block allocated.... > >The processes involved in writing to that partition were using LOTS of >CPU time in system mode. Often top would tell me that over 180% of one >CPU was spent in system mode. > >As you can see, we've solved the problem by doing a little cleaning, >but you may want to look into this for the future.... > > >Thanks for the fast fsck. :-) > > > Roger. > > > Proper solutions from us to your problem include enhanced bitmap scanning code (see 2.4.20 when it comes out), and repacking (see Reiser4.1 when it comes out). -- Hans ----- End forwarded message ----- -- ** R.E.Wolff@BitWizard.nl ** http://www.BitWizard.nl/ ** +31-15-2600998 ** *-- BitWizard writes Linux device drivers for any device you may have! --* * I didn't say it was your fault. I said I was going to blame it on you. *