[net.unix-wizards] Changing window/packet sizes

sam@delftcc.UUCP (Sam Kendall) (05/03/86)

lmjm@icdoc.UUCP says that UKUUCP is hacked to allow one to specify the
window and packet size in L.sys for each connection.  It would be more
interesting, though of course much harder, to have UUCP experiment
during each call to find optimal values for these parameters.  This
would take a new protocol, I guess.

The approach of having UUCP experiment is preferable, first, because it
requires less human parameterization; and second, because it adjusts
better to fluctuating conditions.  For instance, the window and packet
size should go up when there are satellite delays on the line, but they
should go down when there is noise on the line that causes garbling.
U.S.  transcontinental calls vary from call to call both in the
presence/absence of satellite delays (actually, this may only vary
between long-distance companies, not between calls using the same
company) and in the amount of noise.  So it would be nice to have UUCP
compensate dynamically.

Of course, there are other types of connections that favor
different window and packet sizes, with or without automatic
experimentation.  There are half-duplex 9600 baud modems, I have read,
that transmit without errors and simulate full duplex by frequent
handshaking.  There can sometimes be delays of a few seconds in the
transition between send and receive, though.  It seems like a very large
window and packet sizes would make UUCP work quite well with this sort
of modem, so that it could pay for itself if bought for sites that
exchange lots of data long-distance.  Large packet sizes are of course
appropriate for any transmission medium that does its own
error-checking.

Finally, a couple of notes.  (1) Even though it looks like (when you
watch the send/receive lights) that the packet size should be larger,
4.3 UUCP here gets 210 to 220 cps using a Hayes Smartmodem 2400.  Pretty
good.  (2) I'm not actually sure what "window size" is.  If I look like
a total fool in mentioning it above, please forgive me.

----
Sam Kendall			{ ihnp4 | seismo!cmcl2 }!delftcc!sam
Delft Consulting Corp.		ARPA: delftcc!sam@NYU.ARPA