[3803] in APO News
More on apo's death
daemon@ATHENA.MIT.EDU (Ron Peacetree)
Fri Mar 12 13:08:32 1999
To: apo-news@MIT.EDU
Cc: peace@MIT.EDU
Date: Fri, 12 Mar 1999 13:07:46 EST
From: Ron Peacetree <peace@MIT.EDU>
We have had net trouble with apo since we got it. I recently decided
To Do The Right Thing and fix it. I purchased a 3C905B TX to replace
the cheapo Chinese NE2000 clone (!) we had. Improvement was immediate,
but myself, Ira, and Chris noticed that there were too many frame and
RCV errors. Over that last week I've watched that percentage climb
monotonically upward. Last night, we lost the ability to get to the
AFS servers and other net resources. Cold reboots did not help.
Neither did a few other things I tried. Pings from SIPB showed 69% to
72% packet loss. I finally got ahold of another host and plugged it
into our drop. No problem.
The expected pattern in a "bathtub curve" failure of the 3C905 is
different from that observed. The best conclusion is PCI slot related
power problems eating HW. Unfortunately, there are power supply + PCI
slot interaction bugs that could exhibit the same behavior1. The
consensus, and I agree, is complete chasis replacement.
Ron
