| |
PA2AGA > TCPDIG 30.07.97 15:00l 96 Lines 3520 Bytes #-10273 (0) @ EU
BID : 19549_I1VDM
Read: GUEST
Subj: TCP-Group Digest 97/73
Path: DB0RGB<DB0MAK<OK0PKL<OK0PPR<SP6KBL<OK0PBX<OK0PAB<HA5OB<HA1VH<S50BMS<
S50MBR<S50BOX<S50MBL<IV3LAV<IV3AVQ<IV3CJT<IK4ICZ<I4UKI<IK2HDG<IK1MAP<
I1YLM<I1VDM<I1DEP<I1NHK<IK1MSL<IK1ZZN<IK2QCA<IW2FPO<HB9OK<HB9H<HB9OS<
DB0KFB<DB0ZKA<DB0LX<DB0RBS<DB0GE<LX0PAC<ON5VL<ON4RAT<PI8DRS<PI8ZWL<
PI8APD<PI8WNO<PI8GCB<PI8HGL<PI8VNW
Sent: 970729/2047Z @:PI8VNW.#ZH2.NLD.EU #:25376 [Hoek v Holland] FBB5.15c
From: PA2AGA@PI8VNW.#ZH2.NLD.EU
To : TCPDIG@EU
Received: from pa2aga by pi1hvh with SMTP
id AA39240 ; Tue, 29 Jul 97 20:37:19 UTC
Received: from pa2aga by pa2aga (NET/Mac 2.3.64/7.1) with SMTP
id AA00004044 ; Tue, 29 Jul 97 22:30:19 MET
Date: Tue, 29 Jul 97 22:25:48 MET
Message-Id: <tcp_97_73>
From: pa2aga
To: tcp_broadcast@pa2aga
Subject: TCP-Group Digest 97/73
X-BBS-Msg-Type: B
TCP-Group Digest Tue, 29 Jul 97 Volume 97 : Issue 73
Today's Topics:
An experiment?
Send Replies or notes for publication to: <TCP-Group@UCSD.Edu>.
Subscription requests to <TCP-Group-REQUEST@UCSD.Edu>.
Problems you can't solve otherwise to brian@ucsd.edu.
Archives of past issues of the TCP-Group Digest are available
(by FTP only) from ftp.UCSD.Edu in directory "mailarchives".
We trust that readers are intelligent enough to realize that all text
herein consists of personal comments and does not represent the official
policies or positions of any party. Your mileage may vary. So there.
----------------------------------------------------------------------
Date: Mon, 28 Jul 1997 09:25:57 -0700 (PDT)
From: Don Hoffman <hoffman@Eng.Sun.COM>
Subject: An experiment?
From: "Fred R. Goldstein" <fgoldstein@bbn.com>
>
> Of course nothing has changed. TCP's definition was firmed up in the
> 1980s when the Van Jacobson Slow Start algorithm was added, with subsequent
> options not being the norm on the radio. VJSS is an absolute must for
> coping with congestion, as it provides the flexible, automatic backoff
> needed to prevent congestion collapse from occurring. Also TCP's backoff
> on retransmission (separate from its VJSS dynamic window size) is useful for
> shared media, like radio.
>
> AX.25 in "vanilla packet" uses LAP-B procedures, which were developed for
> wireline point-to-point use, where there is no possibility of congestion.
> They are absolutely unworkable otherwise, and packet radio channels
routinely
> display congestion collapse behavior, in which retransmissions cause loss
> which causes retransmissions etc. in a positive feedback loop, and actual
> throughput goes towards nil.
>
> Given this, it's obvious that TCP/IP is better at channel-sharing than
> AX.25. But when you combine the two, the AX.25 users, who do NOT use the
> appropriate backoff behavior, will keep the channel clobbered and cause
> TCP/IP (which will slow down more) to get a disproportionately small share
> of whatever gets through.
>
Yes, quite so.
For a discussion on how TCP reacts to other "TCP-unfriendly" flows
on the same shared links, see Sally Floyd's and Kevin Fall's draft paper at:
ftp://ftp.ee.lbl.gov/papers/collapse.ps
Sally and Kevin mostly focus on router mechanisms to ameliorate the effect,
but they do a pretty good analysis of the effects in the supporting
background. Unfortunately, most of the mechainism they discuss are for
implementation in routers/switches (that drop packets from non-responsive
flows on the input queue to a link), and are not of much help on a shared
channel with distributeed input queueing.
Don Hoffman - KC7YJA
------------------------------
End of TCP-Group Digest V97 #73
******************************
You can send your message for this bulletin
to: tcp-group@pa2aga on .AMPR.ORG-net
or: tcpaga@pi8vnw.#zh2.nld.eu on BBS-net
------
NOT TO: pa2aga@pa2aga or pa2aga@pi8vnw.#zh2.nld.eu PLEASE!!
It will get posted automatically within a few days
Read previous mail | Read next mail
| |
