From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Eckhardt Subject: Re: [9fans] acme mail on unix To: 9fans@cse.psu.edu In-Reply-To: <20060120135038.04867B095F@conchobor.berzerked.org> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-ID: <4947.1137798063.1@piper.nectar.cs.cmu.edu> Date: Fri, 20 Jan 2006 18:01:03 -0500 Message-ID: <4948.1137798063@piper.nectar.cs.cmu.edu> Topicbox-Message-UUID: e2174966-ead0-11e9-9d60-3106f5b1d025 > (the S=2872 is the file size. i'm not sure why stat is so > expensive on their machine.) I suspect it's *not* because of this: AFS directories are (essentially) whole-file transferred to clients who interpret them locally, but each stat() requires an RPC to the server. Instead my guess would be something more like: if you're rebuilding the quota database you can do so by N/M readdir()'s instead of N/M readdir()'s plus N stat()'s, which is probably a huge reduction in disk seeks, which would be attractive since almost everything stat() would be telling you would be irrelevant (e.g., inode number, permissions, actual disk blocks consumed). Dave Eckhardt