[2213] in Athena Bugs
Re: next house rt/ethernet msgs
daemon@ATHENA.MIT.EDU (probe@ATHENA.MIT.EDU)
Tue May 9 07:12:27 1989
From: <probe@ATHENA.MIT.EDU>
Date: Tue, 9 May 89 07:12:11 EDT
To: don@ATHENA.MIT.EDU
Cc: bugs@ATHENA.MIT.EDU, jis@ATHENA.MIT.EDU
In-Reply-To: Donald T. Davis's message of Tue, 9 May 89 00:31:19 EDT,
Reply-To: Richard Basch <probe@ATHENA.MIT.EDU>
I have another algorithm that should get the infrequent case a lot
better and also report a truly disconnected cable, based on the
left-shift history mechanism that we were talking about yesterday.
From your last e-mail message, I did not see anything that will catch
the infrequent case.
If you find the following confusing, talk to me about it later, since I
don't really want to make this e-mail message long.
1 = xmit with shrt
0 = packet received or xmit ok
Only report if:
history contains all 1's or if only the low bit is a 1
Assume 16 bit/packet history:
(16 is arbitrary; choose 32 if you wish)
1111 1111 1111 1111
0000 0000 0000 0001
If it is transient, it will show; if it is steady state, it won't.
Thus, one message may be shown at Next House, but after that, nothing
should appear. The one thing I neglected to mention, if it is not
already obvious is that the history variable should therefore be
initialized to 0.
To the rest of the audience reading bugs: I would argue very highly
against eliminating the messages or making them "debug" only. As is,
the RT's in the cluster provide the only diagnostics of what is going
wrong sometimes. I recall one or two members of ops telling me how they
found a broken cable that took out an entire row in the student center
simply because the RT's decided to complain; the VAXen just sat there
quietly. Also, many of you recall the two problems that we had
experienced on 18.72 because of netboot (VS2000's crash and the ethernet
jam); if it weren't for the RT kernel, we would never have found the
problem as quickly as we did. Something is to be said for keeping these
messages, so personally, I would vote against Jeff's fix.
-Richard
