From owner-freebsd-hackers Sat Dec 20 02:21:05 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.7/8.8.7) id CAA04665 for hackers-outgoing; Sat, 20 Dec 1997 02:21:05 -0800 (PST) (envelope-from owner-freebsd-hackers) Received: from sax.sax.de (sax.sax.de [193.175.26.33]) by hub.freebsd.org (8.8.7/8.8.7) with ESMTP id CAA04650 for ; Sat, 20 Dec 1997 02:20:58 -0800 (PST) (envelope-from j@uriah.heep.sax.de) Received: (from uucp@localhost) by sax.sax.de (8.8.8/8.8.8) with UUCP id LAA11771 for freebsd-hackers@freebsd.org; Sat, 20 Dec 1997 11:20:55 +0100 (CET) (envelope-from j@uriah.heep.sax.de) Received: (from j@localhost) by uriah.heep.sax.de (8.8.8/8.8.5) id LAA29298; Sat, 20 Dec 1997 11:01:26 +0100 (MET) Date: Sat, 20 Dec 1997 11:01:26 +0100 (MET) Message-Id: <199712201001.LAA29298@uriah.heep.sax.de> Mime-Version: 1.0 X-Newsreader: knews 0.9.8 Reply-To: joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch) Organization: Private BSD site, Dresden X-Phone: +49-351-2012 669 X-PGP-Fingerprint: DC 47 E6 E4 FF A6 E9 8F 93 21 E0 7D F9 12 D6 4E References: <19822.882343067@verdi.nethelp.no> <3.0.1.32.19971216232646.00766974@pop.mpc.com.br> From: j@uriah.heep.sax.de (J Wunsch) Subject: Re: 3com 3c509 card X-Original-Newsgroups: local.freebsd.hackers To: freebsd-hackers@freebsd.org Content-Type: text/plain; charset=us-ascii Sender: owner-freebsd-hackers@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Capriotti wrote: > I've heard that some 3com cards have only 3KB buffer; Have any of you > heard of it ? No, i think the smallest possible constellation is 8 KB. The programming documents IMHO offer some possible diversity which you can probe at runtime. The tricky thing with these cards is the buffer space is a FIFO which you have to divide into a Tx and an Rx segment. The actual options how to divide it depend on the total FIFO size, IIRC the 8 KB FIFO could only be divided into 2 KB Tx and 6 KB Rx FIFO. A total of 8 KB only is rather dangerous if your machine is slow, and you're doing NFS. The default NFS block size is 8 KB, and with only 6 KB Rx buffer space, your machine must have fetched the first packet from the buffer before the last one from the NFS server arrived at the NIC (that's ~ 5 ms of time). If you failed to do this, the entire NFS block (consisting of 6 Ethernet packets) was lost, and the server retransmitted it -- with the same problem again. This was a rather widely known NFS deadlock situation back in the days of slower CPUs, and with 3c503's only having 8 KB of buffer space. (You had to mount the server with a smaller blocksize, like 1 KB so each NFS block fits into a single Ethernet packet. That's the ``Slow NFS'' option in sysinstall.) -- cheers, J"org joerg_wunsch@uriah.heep.sax.de -- http://www.sax.de/~joerg/ -- NIC: JW11-RIPE Never trust an operating system you don't have sources for. ;-)