Return to BSD News archive
Newsgroups: comp.os.386bsd.questions
Path: sserve!newshost.anu.edu.au!munnari.oz.au!constellation!convex!convex!cs.utexas.edu!uunet!mcsun!ieunet!ieunet!dec4ie.ieunet.ie!jkh
From: jkh@whisker.lotus.ie (Jordan K. Hubbard)
Subject: Re: ed0: device timeout, freebsd cslip
In-Reply-To: plotkin@alumni.cs.Colorado.EDU's message of Mon, 4 Oct 1993 03: 48:11 GMT
Message-ID: <JKH.93Oct4083135@whisker.lotus.ie>
Sender: usenet@ieunet.ie (USENET News System)
Nntp-Posting-Host: whisker.lotus.ie
Organization: Lotus Development Ireland
References: <J5SHBHSK@math.fu-berlin.de> <1993Oct3.020559.16808@zen.void.oz.au>
<CECt8C.29o@Colorado.EDU>
Date: Mon, 4 Oct 1993 08:31:35 GMT
Lines: 47
OB QUESTION: is anyone else getting simply glacial CSLIP
performance from FreeBSD epsilon? With 386BSD 0.1 I had about
1.1k/sec ftp, with NetBSD 0.9 the lowest I've seen is 1.3k/ sec.
FreeBSD provides a consistent and stunning .6k/sec. Even Linux
pl13 gives .8k/sec!
I get a consistent 1.4 K/sec with my 14.4K FAX modem. Two factors can
hose your slip performance, though it sounds like at least the former
is not your problem but I include it here for completeness:
1. Serial cards with 16450 non-FIFO'ing UARTS. My modem uses the 16550A
chip with FIFO and it makes a major difference.
2. Using the com, rather than the considerably faster sio driver.
I know I'm using the computing equivalent of a 1 inch penis,
but .6k/sec CSLIP with recognized 16550A UARTs, ~5 MIPS and
a perfectly clean phone line? I think I could get better
throughput by using tree sloths as the transport layer.
TST, or Tree Sloth Transport protocol is not supported in FreeBSD
EPSILON, but you can be sure that as soon as the VLL-CPPR (Very Long
Latency - Carrier Pidgeon Packet Relay) standards group gets its act
together regarding TST, we'll be sure and implement it!
Nonetheless, here's the performance I'm getting with FreeBSD EPSILON,
using cslip on an SIO port, right now:
jkh@whisker-> ./tcpblast dec4ie.ieunet.ie 32
11931 bit/s
Or just around 1.491 K/sec. I'd say that's pretty much MAX line
performance for a 14.4K modem! :)
p.s. Hats off to the FreeBSD team for making the kernel source
avaliable separately from all the other spew in the epsilon
release. Nothing like having to ftp 1000 gigabytes of effluvium
at .6k/sec to configure a usable kernel.
Aw shucks.
Jordan
--
(Jordan K. Hubbard) jkh@violet.berkeley.edu, jkh@al.org, jkh@whisker.lotus.ie
I do not speak for Lotus, nor am I even a Lotus employee. I am an independent
contractor.