[73] in Info-AFS_Redistribution
retransmission timeout experiment
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Sat Feb 9 12:07:42 1991
From: Peter Honeyman <honey@citi.umich.edu>
To: info-afs@transarc.com, icg@ifs.umich.edu
Date: Sat, 9 Feb 91 11:37 EST
here are some very preliminary numbers from an overnight experiment
with rx retransmission timeouts over a 9600 baud slip line. the file
server is on the same subnet as one of the slip endpoints; the client
is the other slip endpoint. all the machines are rt/115's running afs
3.0 and the latest ucb networking code.
rto: retransmission timeout
bps: bytes per second
mtu: maximum transmission unit (bytes per slip frame)
the maximum throughput is 867 bps -- one byte requires 10 bit times to
transmit, slip mtu is 576, 20 byte ip header, 8 byte udp, 28 byte rx;
this gives (9600 / 10) * (576 - (20 + 8 + 28)) / 576 = 867.
the results:
rto 2 sec 6 sec 8 sec 12 sec
bps 175 225 350 835
we're also experimenting with dynamic rto estimation. film at 11.
peter
