*BSD News Article 17092


Return to BSD News archive

Newsgroups: comp.os.386bsd.questions
Path: sserve!newshost.anu.edu.au!munnari.oz.au!uunet!emba-news.uvm.edu!sadye.emba.uvm.edu!wollman
From: wollman@sadye.emba.uvm.edu (Garrett Wollman)
Subject: Re: Ethernet [really TCP] performance measurement
Message-ID: <1993Jun13.202515.21826@uvm.edu>
Sender: news@uvm.edu
Organization: University of Vermont, EMBA Computer Facility
References: <hastyC8KHzx.IAv@netcom.com> <1993Jun13.183538.19881@uvm.edu> <hastyC8KsH7.719@netcom.com>
Date: Sun, 13 Jun 1993 20:25:15 GMT
Lines: 38

In article <hastyC8KsH7.719@netcom.com> hasty@netcom.com (Amancio Hasty Jr) writes:
>I am curious about your system configuration your ethernet performance
>seems a bit on the low side.

You're probably comparing it with a 16-bit Ethernet adapter that has a
reasonable amount of memory.  Me no got.  The AT&T EN100 is an 8-bit
adapter (i.e., SLOW SLOW SLOW) with only 16k of memory.  (I am certain
that I could handle 16k TCP windows if only I had 64k of memory...)
In addition to the speed hit that I take from the slow adapter, my
system itself is a 386/20 with no special speed enhancements (it
predates all that by about four years).  Here's what I get on the
loopback (it might be better by now; I'm not sure how long ago this
was done):

ttcp-r: rcvbuf
ttcp-r: accept from 127.0.0.1
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001, sockbufsize=32768  tcp
ttcp-r: 16777216 bytes in 109.09 real seconds = 150.19 KB/sec +++
ttcp-r: 16777216 bytes in 94.63 CPU seconds = 173.14 KB/cpu sec
ttcp-r: 2053 I/O calls, msec/call = 54.41, calls/sec = 18.82
ttcp-r: 0.0user 94.5sys 1:49real 86% 0i+0d 0maxrss 0+0pf 2+1917csw
ttcp-r: buffer address 0xc000

I've got the rudiments of a page-flipping implementation for the
transmit side which should speed things up a little bit, but it won't
help for the receive side.  (Or at least, not until I figure out what
the Right Way of doing it is.)  Currently all I have is iovflipin(); I
don't even know if this works because I never got around to testing
it.  (I think I got around to fixing the mbuf macros not to assume
that everything with an M_EXT necessarily has a cluster allocated...)

-GAWollman

-- 
Garrett A. Wollman   | Shashish is simple, it's discreet, it's brief. ... 
wollman@emba.uvm.edu | Shashish is the bonding of hearts in spite of distance.
uvm-gen!wollman      | It is a bond more powerful than absence.  We like people
UVM disagrees.       | who like Shashish.  - Claude McKenzie + Florent Vollant