[Info-vax] PEA0: Errors

[Christoph Gartmann]

In article <009bf203$0$2993$c3e8da3 at news.astraweb.com>, JF Mezei <jfmezei.spamnot at vaxination.ca> writes:
>
>>> The Alphas all show collisions and late collions where the boot node
>>> has the highest numbers (206 and 120).
>
>
>If the Alphas are in full duplex mode, there should never be collisions.
>By definition, there cannot be a collision in full duplex mode.
>
>Perhaps there are periods of the day when, due to the alignment of the
>moon and the howling of nearby wolves, your alpha secretely renegotiates
>the link to half duplex while nobody is watching ?

No, these counters increase when a crash occurs. The crash is obviously due
to this anti-spam box. The HP switch even reported that it put a link in
half-duplex mode although it has been configured at full duplex. Obviously
the HP switch gets confused when flooded with erronous TCP packets.

>Note that when a node boots, the ethernet interface goes up when power
>is applied and is reset a few times (once during power up when you set
>its parameters, and once or more during boot when VMS takes over and
>sets/resets parameters for the link). Perhaps this is when the errors
>occur.

Usually these numbers are small and remain stable. In our problem case the
numbers increased dramaticially every 36 minutes.

>Are these collision numbers increasing slowly ? Or are they stable once
>you have booted and they no longer increase afterwards ?

Under normal circumstances they are stable. But they increased in steps with a
36 minute delay. Tomorrow we will replace the 3COM switch with the original
one from HP and see, what will happen.

Regards,
   Christoph Gartmann

-- 
 Max-Planck-Institut fuer      Phone   : +49-761-5108-464   Fax: -80464
 Immunbiologie
 Postfach 1169                 Internet: gartmann at immunbio dot mpg dot de
 D-79011  Freiburg, Germany
               http://www.immunbio.mpg.de/home/menue.html