From owner-freebsd-hackers Thu Jun 12 13:21:28 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.5/8.8.5) id NAA18796 for hackers-outgoing; Thu, 12 Jun 1997 13:21:28 -0700 (PDT) Received: from verdi.nethelp.no (verdi.nethelp.no [195.1.171.130]) by hub.freebsd.org (8.8.5/8.8.5) with SMTP id NAA18791 for ; Thu, 12 Jun 1997 13:21:21 -0700 (PDT) From: sthaug@nethelp.no Received: (qmail 1676 invoked by uid 1001); 12 Jun 1997 20:21:16 +0000 (GMT) To: hackers@freebsd.org Subject: New Netperf throughput numbers for FreeBSD X-Mailer: Mew version 1.05+ on Emacs 19.28.2 Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Date: Thu, 12 Jun 1997 22:21:16 +0200 Message-ID: <1674.866146876@verdi.nethelp.no> Sender: owner-hackers@freebsd.org X-Loop: FreeBSD.org Precedence: bulk I just submitted a new set of throughput numbers for Fast Ethernet and FreeBSD to Rick Jones, the Netperf maintainer. In short, we can fill the wire :-) I measured 93.57 Mbit/s with Netperf. This was the best of several measurements with different socket buffer sizes and read/write sizes. However, *all* of the measurements were above 93 Mbit/s. If you look at the bits on the wire, 93.57 Mbit/s (application to application) corresponds to 93.57 * (1538/1440) = 99.94 Mbit/s on the wire. Note that 1440 is the correct number, not 1460 - because I made no changes to the default use of RFC 1323 and RFC 1644 extensions (so each packet has 20 bytes of TCP options). The setup was: Sender: noname machine, PPro-200 with 256 KB cache, 440FX chipset, BCM Advanced Research SQ600 mainboard, 64 MB memory. Kingston (DEC 21140 based) 100BaseTX network card. FreeBSD 2.2-BETA operating system. Receover: noname machine, P-133 with 256 KB cache, 430VX chipset, QDI P5I430VX motherboard, 32 MB memory. Intel Pro 100/B 100BaseTX network card. FreeBSD 3.0-970124-SNAP operating system. Both machines were run with no other load during the test. They were connected via a Cisco Catalyst 5000 switch. Connections to both hosts were full duplex. The network was isolated. The P-133 was clearly a limiting factor. Running 'top' showed that it was spending something like 99% of the time in kernel or interrupt mode during this test - so even with a network faster than 100 Mbps Ethernet, *this* particular setup wouldn't run much faster. However, the PPro-200 still had plenty of CPU left. Steinar Haug, Nethelp consulting, sthaug@nethelp.no