From owner-freebsd-questions  Thu Aug 22 21:44:17 1996
Return-Path: owner-questions
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id VAA27600
          for questions-outgoing; Thu, 22 Aug 1996 21:44:17 -0700 (PDT)
Received: from news.interworld.net (news.interworld.net [206.124.224.6])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id VAA27593
          for <questions@freebsd.org>; Thu, 22 Aug 1996 21:44:13 -0700 (PDT)
Received: (from pete@localhost) by news.interworld.net (8.7.5/8.7.3) id VAA22753; Thu, 22 Aug 1996 21:44:10 -0700 (PDT)
Date: Thu, 22 Aug 1996 21:44:10 -0700 (PDT)
From: Peter Carah <pete@news.interworld.net>
Message-Id: <199608230444.VAA22753@news.interworld.net>
To: questions@freebsd.org
Subject: Re: Samba file I/O performance
In-Reply-To: <199608051230.MAA04235@gamespot.com>
Organization: InterWorld Communications
Sender: owner-questions@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

In article <199608051230.MAA04235@gamespot.com> ian writes:

>Which ethernet cards are you using and which ones were running afoul?
>(fbsd & win95)
>

As a slight aside, and for a data point, I am using (at another company :-)
samba to play real-time mpeg video through; the server is an SS1000E with
an HME ethernet card (100-base-T) (actually this is the cute SunSwift card
with 100-base-T and wide scsi on the same single-format Sbus card) and the 
W95 machine has (I think) an Intel 100-b-T card (and a custom decoder
card).  This will play (at least) 5mbit mpeg-2 streams (with interleaved
full AC3 audio for another half mbit) with no hiccups.  (the streams are
recorded on a raid disk set using the wide scsi on the same card... 
iozone reports 6.5 or so mbytes read rate (and about 4 write due to the
raid-5 parity stripe) for the scsi+raid combo to the server.  Samba is
slow compared to that but still OK.)

The card-buffer overrun problem is famous; with NFS and a 2k buffer card
(like WD/SMC 8003E, etc) you get *NO* throughput unless you mount with
rsize=wsize=1024.  (NFS retries the whole 8k chunk so the second packet
will always overrun the card and the whole thing never makes it.  This
fails independent of MSS/MTU/MRU.)  (this is presuming the server is
typically fast - both Sun4 and SGI normally are plenty fast enough to do
in fbsd machines with WD/SMC8003 cards.)

-- Pete