From owner-freebsd-current Fri Feb 24 10:51:50 1995 Return-Path: current-owner Received: (from majordom@localhost) by freefall.cdrom.com (8.6.9/8.6.6) id KAA22116 for current-outgoing; Fri, 24 Feb 1995 10:51:50 -0800 Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.34]) by freefall.cdrom.com (8.6.9/8.6.6) with ESMTP id KAA22110 for ; Fri, 24 Feb 1995 10:51:44 -0800 Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id FAA02024; Sat, 25 Feb 1995 05:39:45 +1100 Date: Sat, 25 Feb 1995 05:39:45 +1100 From: Bruce Evans Message-Id: <199502241839.FAA02024@godzilla.zeta.org.au> To: roberto@blaise.ibp.fr, starkhome!gene@sbstark.cs.sunysb.edu Subject: Re: Decrease in namei cache effectiveness? Cc: current@FreeBSD.org Sender: current-owner@FreeBSD.org Precedence: bulk >> starting a make world. I stopped it after I noticed that the disk was getting >> an incredible pounding during C compilations that previously would be >> almost totally CPU-bound. My impression from "systat -vmstat" was that >> somehow the effectiveness of the namei cache had gotten drastically reduced, >> and the lookups of the names of include files were causing a lot of disk >> activity. >> >> Has anyone else seen this? >Same here. Very very slow (I was compiling sendmail, guess why ? :-)) Same here. `du -a /var | wc' accesses the disk every time to stat only 68 files. I suspect revision 1.30 to vfs_bio.c (1995/02/22 09:30:13) broke something. It was supposed to _stop_ the buffers associated with directories and metadata being thrashed by regular i/o. My getnewvnode() stops some of this thrashing (and thrashing of file data) by preferring to reuse vnodes with no attached buffers. Before 1995/02/22, this increased the maximum amount of directory and metadata info by a factor of about 10. I want it to be increased by another factor of 10. Bruce