From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hans Reiser Subject: Re: Silly question, defrag Date: Fri, 05 Apr 2002 19:07:36 +0400 Message-ID: <3CADBDB8.3000809@namesys.com> References: Mime-Version: 1.0 Content-Transfer-Encoding: 7bit list-help: list-unsubscribe: list-post: List-Id: Content-Type: text/plain; charset="us-ascii"; format="flowed" To: Benjamin Scott Cc: ReiserFS List Benjamin Scott wrote: >On Fri, 5 Apr 2002, at 12:31am, Ross Vandegrift wrote: > >>>I think it would seem likely that, once a filesystem does become fragmented, >>>performance is going to suck mud. >>> >>I'm curious though - shouldn't the elevator read algorithms take care of >>this, for the most part? If you wait to do a read until you have a few, >>and read sequentially across the disk, you won't be hit with any seeks. >> > > I am no expert, but I suspect there is only so much the elevator can do. >For a moderately small read, I image your scenario would work. But given a >larger read and enough fragmentation, you are going to have to seek all over >the place. Sure, you could increase the wait time in the elevator, and use >tons of buffers to re-order reads, but you will eventually start spending so >many resources on the elevator that you are back to a performance problem >again. > Actually, the larger the read, the better the elevator can compensate. The problem is that if you do a whole bunch of separate reads of a whole bunch of files, and you wait for each read to complete before doing the next one, the elevator cannot do a good job. Also, if you have data you don't want to read interleaved with what you do want to read, then your transfer rate will be decreased. Hans