Date: Sun, 9 Nov 1997 08:16:55 -0800 (PST) From: rtm@eecs.harvard.edu To: freebsd-gnats-submit@FreeBSD.ORG Subject: kern/4992: SCSI disk scheduling disabled in 2.2.5 Message-ID: <199711091616.IAA20217@hub.freebsd.org> Resent-Message-ID: <199711091620.IAA20371@hub.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 4992 >Category: kern >Synopsis: SCSI disk scheduling disabled in 2.2.5 >Confidential: no >Severity: non-critical >Priority: low >Responsible: freebsd-bugs >State: open >Class: sw-bug >Submitter-Id: current-users >Arrival-Date: Sun Nov 9 08:20:00 PST 1997 >Last-Modified: >Originator: Robert Morris >Organization: Harvard University >Release: 2.2.5 >Environment: FreeBSD frenulum.eecs.harvard.edu 2.2.5-RELEASE FreeBSD 2.2.5-RELEASE #1: Sat Nov 8 17:45:42 EST 1997 rtm@frenulum.eecs.harvard.edu:/amd/dominator/disk/glan25/kung/rtm/sys-2.2.5/compile/BOTH i386 >Description: The call to tqdisksort() in sd.c seems to be effectively commented out by an #ifdef SDDISKSORT. If one turns on SDDISKSORT, tqdisksort() still doesn't schedule read requests, because of the "goto insert" early in the function. Fixing all this improves my system's total throughput on heavy multi-user loads by about 30% (when 20 processes are doing small reads from different parts of the disk). With this load, scheduling alone does slightly better than AHC_TAGENABLE by itself. Both together improve performance by an additional 15%. >How-To-Repeat: >Fix: >Audit-Trail: >Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199711091616.IAA20217>