[19] in Info-AFS_Redistribution
Re: Oh, btw.
daemon@ATHENA.MIT.EDU (Michael T. Stolarchuk)
Fri Nov 16 12:02:26 1990
To: nydick@psc.edu
Cc: mts@terminator.cc.umich.edu, Info-AFS@transarc.com
In-Reply-To: Your message of Fri, 16 Nov 90 09:09:49 -0500.
Date: Fri, 16 Nov 90 11:57:24 -0500
From: Michael T. Stolarchuk <mts@terminator.cc.umich.edu>
Thats interesting. What are you trying to find out?
We have a different version of rxgen which allows the contents
of each of the operations to be recorded. We make a log from
that and use post porcessing tools to try and answer questions
about the protocol on the wire. The log is written compressed to
cut down on the number of bits needed to represent the data.
Honeyman did the compression and recording part. I did the alternate
version of rxgen. Our objects included getting info on the
number of operations being targetted to servers, the time
the operations take, and the resources used for each (like
cpu time, disk I/O's, etc). We have seen some odd values for
some of the numbers, and are trying to figure out what the values
mean.
We know some of the numbers are odd because to two other tools we have.
The first is a set of mods for sniffers' to be able to display
rx and higher protocols. This helps since the sniffer does a really
good job of listening to the wire. It makes sure we get all the data.
But the sniffer ain't so good at listening for long periods of time.
We have another tool which reads the ethernet on IBM/RT's... it also
understands the rx and higher protocols. There is a version of that
tool which we use to cound 'ethernet' number of different operations, and
to also listen for the start and finish of calls. That way we can
get average time for the operations, and standard deviation...
We use the tools against each other to look for areas where either
has errors... not necessarily bugs, but errors.
mts. the snail.
