home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
kermit.columbia.edu
/
kermit.columbia.edu.tar
/
kermit.columbia.edu
/
newsgroups
/
misc.19980424-19980901
/
000317_news@newsmaster….columbia.edu _Thu Jul 30 12:05:36 1998.msg
< prev
next >
Wrap
Internet Message Format
|
1998-08-31
|
5KB
Return-Path: <news@newsmaster.cc.columbia.edu>
Received: from newsmaster.cc.columbia.edu (newsmaster.cc.columbia.edu [128.59.35.30])
by watsun.cc.columbia.edu (8.8.5/8.8.5) with ESMTP id MAA16378
for <kermit.misc@watsun.cc.columbia.edu>; Thu, 30 Jul 1998 12:05:36 -0400 (EDT)
Received: (from news@localhost)
by newsmaster.cc.columbia.edu (8.8.5/8.8.5) id MAA12010
for kermit.misc@watsun; Thu, 30 Jul 1998 12:05:35 -0400 (EDT)
Path: news.columbia.edu!watsun.cc.columbia.edu!fdc
From: fdc@watsun.cc.columbia.edu (Frank da Cruz)
Newsgroups: comp.unix.sco.misc,comp.protocols.kermit.misc
Subject: Re: Xenix to 5.0.4 Data Transfer ? Stay with UUCP
Date: 30 Jul 1998 16:05:33 GMT
Organization: Columbia University
Lines: 74
Message-ID: <6pq5kd$i65$1@apakabar.cc.columbia.edu>
References: <35bbffb0.3086687@news.ricochet.net> <35bb532d.1519852@news.ricochet.net> <Ewu3vs.1vy@bokonon.ussinc.com> <35c1f9d4.5908262@news.ricochet.net>
NNTP-Posting-Host: watsun.cc.columbia.edu
Xref: news.columbia.edu comp.unix.sco.misc:72130 comp.protocols.kermit.misc:9027
In article <35c1f9d4.5908262@news.ricochet.net>,
Jeff Liebermann <jeffl@comix.santa-cruz.ca.us> wrote:
: On Wed, 29 Jul 1998 02:21:27 GMT, stephen@bokonon.ussinc.com
: (Stephen M. Dunn) wrote:
:
: >$Kermit, zmodem etc file transfer performance:
: >$ http://www.columbia.edu/kermit/perf.html
: >
: >Is this the comparison that Chuck Forsberg (the guy who invented
: >Zmodem) declared was garbage?
:
There was a year-long debate over the Kermit benchmarks in
1993-94, which can be exhumed by anybody who keeps newsgroups that
far back -- comp.dcom.modems was the main one, I think; maybe
even comp.protocols.kermit, before it was split into .announce.
and .misc.
In 1993, the general public believed the party line that Kermit
protocol was instrinsically slow, and "should not be used except
as a last resort". This was stated in numerous sources,
including telecom dictionaries and modem manuals (in fact, USR
manuals still say this). The benchmarks were to demonstrate
that Kermit protocol was not intrinsically slow, and to show
where the common misperceptions came from (primarily, minimal
Kermit implementations in 3rd-party software packages).
: I went through this discussion with Chuck many years ago. As I
: vaguely recall, I wasn't all that convinced by his arguements
: either. I did a comparision for myself about 4 years ago and
: found that with buffered UART's, proper tweaking and decent
: modems, I can get almost identical performance out of large
: packet UUCP, Zmodem and Kermit.
:
On a good connection. On a poor connection, note that Kermit
uses selective packet retransmission for recovery, while Zmodem
rewinds to the point of error and starts over, which can be a
rather large hit when there is a lot of latency in the connection.
: The difference is that Zmodem figured out the optimum packet
: size without any configuration or tweaking while UUCP and
: Kermit required parameter juggling.
:
Kermit is a much more pessimistic protocol than Zmodem. By
default, it *assumes* a poor connection -- lack of transparency,
lack of effective flow control, etc, and therefore is able to
usually work out-of-the-box on these connections, whereas Zmodem
needs parameter juggling (window size, escaping, etc) to work at
all on the same connections. That's the tradeoff -- in a perfect
world, of course, there would not need to be tradeoffs.
On the other hand, when you tell Kermit to use a particular
packet size (any length up to 9024 is permitted), it treats that
as a maximum. Modern Kermits use a slow-start algorithm to test
the waters, and when there are errors, uses exponential backoff
with slow regrowth to recover performance in case the errors
were transitory. Both packet lengths and timeouts are totally
dynamic and constantly adjust themselves to the connection. You
can watch these algorithms in action on the C-Kermit 6.0 or
Kermit 95 fullscreen file-transfer display.
The point about the 24K binary file travelling at
hyper-lightspeed is somewhat valid, and in retrospect I should
have chosen a longer file (the point of this measurement was to
dramatically illustrate the run-length compression feature).
The results reported were as shown by the software timers, but
in those days we had integer timers with 1-second granularity,
and so for short transfers at high speeds, the rouding error
could be rather significant. Nowadays we have floating point
timers and much-more accurate time and speed reports, e.g. in
C-Kermit 6.1 (presently in Beta test):
http://www.columbia.edu/kermit/ck61.html
- Frank