Figured someone here might be interested in some of this thread, so I'm
cross posting it.. Anyone got any feedback?
-- Blue Lang IBM Global Services P: (919) 486-5183 E: wdlang@us.ibm.com ---------- Forwarded message ---------- Date: Mon, 6 Sep 1999 14:20:21 +0200 From: Russell Coker <russell@coker.com.au> To: Matthew Kirkwood <weejock@ferret.lmh.ox.ac.uk> Cc: reiserfs@devlinux.com Subject: (reiserfs) Re: files with numeric names >> Assertion: Squid and INN have file access patterns that are different >> from all other applications. > >Is this really true? I would have thought that they just pushed the >filesystem a lot harder. Squid: average file size ~13K. INN: average file size <4K. My root file system (includes a small squid cache): average file size 25K. I consider that alone to be convincing evidence that INN and Squid have different access patterns. >> Assertion: a file system could recognise the files used by such >> programs by their numeric file names (INN) and hexadecimal file names >> (Squid). > >I think you'd get a surprising number of false positives this way. Do any examples come to mind? >> Theory: a file system could change the way it handles meta-data when >> it sees such file names to improve performance of these applications. > >In what ways would its behaviour change? Cache directories and other meta-data agressively to the exclusion of caching file data. ATIME resolution changes. Probably more possibilities exist, I just can't think of them at the moment. >Do bear in mind that anybody doing serious squid or INN will have >separate filesystem (on separate spindles) for the purpose. You >may not think that this should have to be the case, but I think that >it's a lot nicer than "magic" behaviour. True. Maybe some of these things could be done better by mount options. -- I'm in Utrecht. I'd like to meet any Linux users in the area, or any other part of the Netherlands.Received on Mon Sep 06 1999 - 12:55:01 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:16 MST