The TRUTH about KERMIT NEWS
- or -
What Columbia University doesn't
know about File Transfers
may surprise you!
Last July Columbia University used its non-profit mail permit to
distribute thousands of copies of Kermit News Number 5. This
issue starts with an announcement that Columbia now accepts
Credit Card orders. Under the heading "Kermit Saves You Money"
Columbia invites businesses to switch from other vendors'
shareware and commercial hardware to their product. Using
a non profit mail permit to distribute this advertising gives
Columbia a crucial advantage over tax paying competitors.
I particularly object to Columbia's publishing incorrect and
misleading information about the performance of file transfer
software. That Columbia does so at public expense is an added
In "The Truth about Kermit File Transfer Performance" tenured
Columbia University professor Frank da Cruz proudly announces
that Kermit "outperforms X-, Y-, and ZMODEM protocol transfers
every time." (Kermit News July 1993 p.14) To lend credence to
this incredible assertion, Frank includes results of so-called
Frank's benchmarks and interpretation leave much to be desired.
It is not my intention to defend a competitor's product. But,
the reported Kermit efficiency of Procomm Plus (93%) and Zstem
(97%) are very close to that of Columbia's own MSKermit (99%).
Frank's 104% figure for MSKermit is misleading because the
reduced control character prefixing responsible for the MSKermit
speedup also works sending to Procomm Plus. Columbia does not
disclose this inconvenient result. In fact, C-Kermit sending to
Procomm Plus transferred some of Columbia's test files in less
time than the C-Kermit-MSKermit combination required.
After shading the truth about competitors' Kermit downloading
performance, Frank's paper proceeds to redefine the YMODEM
protocol I developed in the 1980's. YMODEM is a batch protocol
based on XMODEM that uses 128 or 1024 byte data blocks. The
sz/sx/sb 3.24 Program that Columbia used for its "tests" also
supports YMODEM-g, a high speed streaming version of YMODEM.
All of this is explained in the sz documentation, yet the
protocol experts at Columbia University were clueless as to the
actual performance potential of the sz 3.24 software they used
for their True-Life Benchmarks.
Thanks to Columbia's ignorance of file transfer protocols I
don't know what protocol they actually used for the "XMODEM" and
"YMODEM" tests listed in Columbia's True-Life Benchmarks. But
Columbia's published numbers reveal a seriously slovenly
experimental procedure. XMODEM and ZMODEM in all their forms
transmit file data verbatim, without any prefixing. As a
result, data patterns have no effect on X/YMODEM throughput.
(Insert Chart 1, ys.gif, "Throughput vs File Size")
This is clearly shown the chart above, where true XMODEM/YMODEM
throughput is affected only by the file length. (Start-up time
is less important for longer files.)
(Insert Charts 2a and 2b, ks.gif and ks2.gif)
Compare this with Columbia's data. These variations, which are
not a function of file size, show an experimental error of more
than 20 per cent.
(Insert chart #3 fudge.gif)
In addition, Columbia Kermit programs significantly understate
the time required to perform a file transfer. Kermit reports a
file transfer took 112.98 seconds when in fact the machines were
tied up more than nine seconds longer. Professional-YAM's
ZMODEM self reported transfer time was much more accurate,
understating the actual time by less than .5 second. When this
issue was raised on the Usenet comp.dcom.modems newsgroup, the
frogheads suggested we ignore this Kermit overhead. To avoid
such bias, the actual time should be measured with the Unix
"time" command and verified with a stopwatch.
One set of Columbia's measurements was so preposterous that I
simply couldn't resist having a bit of fun with it. We noticed
that cutting the communications speed from 19200 to 9600
actually increased the speed of a Columbia Kermit file transfer
from 2648 to 2736 characters per second. (p.13, Tables 2 and
5.) This is the moral equivalent of putting a 30 mph governor on
your car only to find highway travel is now quicker than when
you were pushing 60. When I announced the "Columbia Stochastic
Telepathic Kermit Hyperprotocol" to "explain" Columbia's numbers
the Kermit faithful quite lost their sense of humor. The
frogheads have flamed about tis "news release" but they have yet
to explain how Columbia came up with these crazy numbers.
And so we come to the The 1994 Protocol Shoot-out of Jan 6 1994
held at the Omen Technology World Corporate Headquarters.
Unlike Columbia's Kermit News True-Life Benchmarks, the
Shoot-out was held in public. Frank da Cruz was invited to
attend, to verify the honesty and competence of the tests. As
promised, we did not benchmark competitors' software while the
machine was overloaded with other work.
One of the tricks Columbia used to discredit ZMODEM in their
"True-Life Benchmarks" was to carefully select files with
redundant data that responds unusually well to Kermit's run
length compression while at the same time refusing to use ZMODEM
The most egregious instance of this deception is seen with
Columbia's selection of a Sun Sparc binary of the "uuencode"
program. This is a tiny program stored in a 24576 byte file.
All but a few thousand are nulls. Naturally this does wonders
for Kermit transfer speeds. Had Columbia's protocol boffins
read the sz 3.24 documentation they could have discovered that
ZMODEM compression does even better. Kermit gets 320 per cent
efficiency on this most extraordinary file; ZMODEM compression
gets an even better 363 per cent efficiency. A separate
compression program does even better at 646 per cent, twice as
good as Kermit.
Real world users download mostly compressed files, and here the
winner is YMODEM-g, closely followed by MobyTurbo(Tm) ZMODEM,
regular ZMODEM, with Kermit bringing up the rear.
(Insert Chart # gifspeed.gif)
So why is Kermit not the fastest when it only prefixes a few
control characters? The question is incorrect. Columbia claims
that only 0, 1, and 129 need to be prefixed, but this doesn't
work sending to MSKermit 3.11. The recommendations made by
Frank da Cruz do work:
SET CONTROL UNPREFIX ALL
SET CONTROL PREFIX 1 13 129 141 ; For sending to MS-DOS Kermit
This makes C-Kermit prefix 8 characters, 0, 1, d, 7e, 81, 8d,
a3, and fe (hex). Knowing this, one can understand why YMODEM-g
is fastest since it does not prefix any control characters.
MobyTurbo(Tm) ZMODEM is next with one control character
prefixed, followed by standard ZMODEM which prefixes 5 control
characters (7 when dialing out). Kermit not only prefixes 8
characters, but suffers from the startup and shutdown delays
The most impressive demonstration, and the likely reason Frank
da Cruz did not choose to appear at the Shootout, is Kermit's
performance in the noise test. Or, to be more precise, Kermit's
non-performance. In the Kermit News True-Life Benchmarks, Frank
da Cruz specified a window size of 5 and a packet length of
5000. Undoubtedly Frank chose this long packet length to
minimize Kermit's high per packet overhead compared to ZMODEM SUBpackets.
XMODEM, YMODEM and Kermit cannot resend a packet with a
different size. The problem is even worse with Kermit's
selective retransmission protocol. When we attempted to
replicate Columbia's 9600 bps plus noise test (Table 5) Kermit
failed every time. Well, almost every time. The first few
times Kermit worked, but that was only because C-Kermit quietly
refused to send the specified 5000 byte packets because a needed
"set buffer" command was not included. When Kermit actually
uses the specified 5000 byte packets with noise bursts at 2
second intervals, no data packets get through. Why? It takes 5
seconds to send a 5000 byte packet at 9600 bps but the specified
noise bursts come faster than that. Kermit croaks every time unless
the Jim Kirk Kobiashi Maru procedure is used (relax the test).
I made an informal survey of the guests at the shootout. As so
many have noted, line noise today manifests itself mostly in
disconnected calls. When this happens ZMODEM's Crash Recovery
is relevant, allowing the safe resumption of interrupted file
transfers. Since Kermit has no such feature Columbia chose not
to discuss this most vital of ZMODEM's many features.
What Columbia University's Kermit mavens don't know
about file transfer protocols may hurt you.
Chuck Forsberg WA7KGX caf@omen.COM 503-621-3406
Author of YMODEM, ZMODEM, Professional-YAM, ZCOMM, and DSZ
Omen Technology Inc "The High Reliability Software"
TeleGodzilla BBS:621-3746 FAX:621-3735 CIS:70007,2304
Genie:CAF 17505-V NW Sauvie IS RD Portland OR 97231