From owner-freebsd-hackers  Fri Mar  1 01:14:32 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id BAA27925
          for hackers-outgoing; Fri, 1 Mar 1996 01:14:32 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id BAA27918
          for <hackers@freebsd.org>; Fri, 1 Mar 1996 01:14:26 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id UAA04513; Fri, 1 Mar 1996 20:10:20 +1100
Date: Fri, 1 Mar 1996 20:10:20 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199603010910.UAA04513@godzilla.zeta.org.au>
To: davidg@Root.COM, jgreco@brasil.moneng.mei.com
Subject: Re: 100-baseT hub recommendation?
Cc: angio@aros.net, hackers@freebsd.org, jc@irbs.com
Sender: owner-hackers@freebsd.org
Precedence: bulk

[SMC Etherpower 10/100's vs Intel 100baseT]

>   There aren't any known bugs in the driver. Performance is about 5% higher
>with the Intel card compared to the SMC/DEC.

There seems to be a latency problem in the de driver or hardware.  The ed
driver gives about 3% higher performance than the de driver with my 10Mbps
cards:

machine a: P133 - ASUS P55TP4XE - Accton EN1203 (DEC 21040)
machine b: 486DX2/66 - No-name slow ISA - No-name AE-200JL (NE2000/WD8013EBT clone)
machine c: 486DX/33 - No-name slow ISA - No-name AE-200JL (NE2000/WD8013EBT clone)

a -> c as seen be c
ttcp-r: socket
ttcp-r: accept from 192.168.2.1
ttcp-r: buflen=8192, nbuf=2048, align=16384/+0, port=5001  tcp
ttcp-r: 167772160 bytes in 154.38 real seconds = 1061.26 KB/sec +++
ttcp-r: 116133 I/O calls, msec/call = 1.36, calls/sec = 752.24
ttcp-r: 1.3user 32.9sys 2:34real 22% 16i+228d 228maxrss 0+2pf 115572+505csw

b -> c as seen by c
ttcp-r: socket
ttcp-r: accept from 192.68.2.2
ttcp-r: buflen=8192, nbuf=2048, align=16384/+0, port=5001  tcp
ttcp-r: 167772160 bytes in 149.82 real seconds = 1093.55 KB/sec +++
ttcp-r: 116367 I/O calls, msec/call = 1.32, calls/sec = 776.69
ttcp-r: 1.3user 33.3sys 2:29real 23% 16i+226d 220maxrss 0+2pf 116075+502csw

As usual, PIO gives better performance than DMA :-).  The busmastering
controller has less overhead of course:

machine a:  8% Sys +  4% Intr = 12%
machine b: 25% Sys + 55% Intr = 80%

Bruce