*BSD News Article 4955


Return to BSD News archive

Newsgroups: comp.unix.bsd
Path: sserve!manuel!munnari.oz.au!uunet!usc!sdd.hp.com!network.ucsd.edu!qualcom.qualcomm.com!servo.qualcomm.com!karn
From: karn@servo.qualcomm.com (Phil Karn)
Subject: Re: [386BSD]
Message-ID: <1992Sep13.061916.27672@qualcomm.com>
Sender: news@qualcomm.com
Nntp-Posting-Host: servo.qualcomm.com
Organization: Qualcomm, Inc
References: <4ebZONa00WB3Qzp15W@andrew.cmu.edu> <1992Aug29.163711.25695@NeoSoft.com>
Date: Sun, 13 Sep 1992 06:19:16 GMT
Lines: 22

In article <1992Aug29.163711.25695@NeoSoft.com> karl@NeoSoft.com (Karl Lehenbauer) writes:
>Unless you want to do ham packet radio stuff, the Berkeley TCP/IP and SLIP
>implementations that come with 386BSD are superior to KA9Q, which was
>designed for non-multitasking DOS operation and, thus, incurs some serious
>penalties when running under Unix.

Ahem.

Although I agree that my stuff isn't designed to run efficiently under
UNIX, I beg to differ about the "quality" of the implementation. I've
long had all the same stuff in my TCP/IP and SLIP that Berkeley has:
VJ header compression, VJ srtt/mdev rto calculation, clamped
retransmission backoffs, slow start, fast recovery, Nagle tinygram
avoidance, etc. It's fast enough to never be the bottleneck on a FTP
(as opposed to the wire or the disk.)  And it has avoided many of the
problems that still nag the descendants of some previous Berkeley
TCPs, like closing all TCP connections to a given host whenever you
receive ANY ICMP message that references that host...

Phil