[5] in 6.033-lab
Sequential Request ID problems
daemon@ATHENA.MIT.EDU (Constantine P. Sapuntzakis)
Thu Mar 6 23:45:25 1997
To: 6.033-lab@MIT.EDU
Date: Thu, 06 Mar 1997 23:46:05 EST
From: "Constantine P. Sapuntzakis" <csapuntz@MIT.EDU>
Some of you have picked sequentially increasing Request IDs as a way
of uniquely tagging requests from the client to the server. You have
argued that it is only necessary then store at the server the Request ID of the
current request. Packets from older requests can be detected by
low sequence numbers.
This approach does NOT work. The problem is that there is no easy way to
maintain monotonically increasing in two clients on a given machine/port
where one program run the other) but coincidentally listening on the
same machine/port. Also, in the case of a client crash, all information
about the previous sequence numbers is lost and a new one has to be
created. We do not want the new one to be rejected just because it is
smaller than some older Request ID.
So the server probably shouldn't use numerical comparisons other than equality
on request IDs.
-Costa
caspuntz@mit.edu
