From owner-cvs-all Fri Nov 6 12:16:25 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id MAA25476 for cvs-all-outgoing; Fri, 6 Nov 1998 12:16:25 -0800 (PST) (envelope-from owner-cvs-all@FreeBSD.ORG) Received: from niobe.ewox.org (ppp114.uio.no [129.240.240.119]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id MAA25300 for ; Fri, 6 Nov 1998 12:15:57 -0800 (PST) (envelope-from finrod@niobe.ewox.org) Received: (from finrod@localhost) by niobe.ewox.org (8.9.1/8.9.1) id TAA00414; Fri, 6 Nov 1998 19:29:39 +0100 (CET) (envelope-from finrod) To: committers@FreeBSD.ORG Subject: sendfile performance Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 8bit From: finrod@ewox.org (Dag-Erling Coidan Smørgrav) Date: 06 Nov 1998 19:29:38 +0100 Message-ID: <86vhks4qul.fsf@niobe.ewox.org> Lines: 45 X-Mailer: Gnus v5.3/Emacs 19.34 Sender: owner-cvs-all@FreeBSD.ORG Precedence: bulk Here are some numbers for sendfile performance on 100 Mbps Ethernet (niobe and luna are connected back-to-back with a crossover cable): finrod@niobe ~/src/sft$ ./tcpsendfile -1 luna 5400 /kernel -r100 count = 100 finrod@niobe ~/src/sft$ /usr/bin/ti./tcpsendfile -1 luna 5400 /kernel -r100 tic time tip finrod@niobe ~/src/sft$ /usr/bin/time ./tcpsendfile -1 luna 5400 /kernel -r100 count = 100 20.51 real 0.04 user 1.09 sys finrod@niobe ~/src/sft$ /usr/bin/time ./tcpsendfile -2 luna 5400 /kernel -r100 count = 2300 21.70 real 0.04 user 6.93 sys finrod@niobe ~/src/sft$ /usr/bin/time ./tcpsendfile -1 luna 5400 /kernel -r100 -b8192 count = 18200 21.16 real 0.07 user 1.34 sys finrod@niobe ~/src/sft$ /usr/bin/time ./tcpsendfile -2 luna 5400 /kernel -r100 -b8192 count = 18200 20.97 real 0.11 user 7.30 sys finrod@niobe ~/src/sft$ /usr/bin/time ./tcpsendfile -1 luna 5400 /kernel -r100 -b1024 ^C/usr/bin/time: command terminated abnormally 39.97 real 9.18 user 29.10 sys I interrupted the last transfer because netstat showed absolutely no activity on the wire. During the other tests I got a quite satisfying throughput: root@niobe /home/finrod# netstat -i 1 input (Total) output packets errs bytes packets errs bytes colls 1774 0 107580 4964 0 7598766 0 1775 0 107520 4967 0 7591196 0 1777 0 107400 4973 0 7598766 0 1772 0 107580 4964 0 7603308 0 1776 0 107580 4973 0 7598766 0 1773 0 107640 4966 0 7601794 0 1778 0 107460 4974 0 7589998 0 In other words, the wire was maxed out. Perhaps this discussion should be moved to -hackers? DES -- Dag-Erling Smørgrav - finrod@ewox.org To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message