Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!sdd.hp.com!hp-pcd!hpfcso!hpfcmgw!glen
From: glen@hpfcmgw.HP.COM (Glen Robinson)
Newsgroups: comp.sys.hp
Subject: Re: 9000/370 Problems...
Message-ID: <1080162@hpfcmgw.HP.COM>
Date: 26 Jul 90 17:49:15 GMT
References: <13484@udenva.cair.du.edu>
Organization: HP Fort Collins, CO
Lines: 43

The H-P field guys keep insisting that it is a power problem because that
is the most likely cause of parity panics on a 370.  Parity can ONLY occur
on a read from memory (that is the only time it is checked and this is
done by hardware not software).  Therefore the problem could have only
happened during the previous write to the location or at some time after
the initial write.  Problems that occur duuring writes to a memory location
can quite reliably be found with memory diagnostics including the boot time
memory diagnostic.

Problems occuring after a cell is written usually are caused by one of two
things:
    1.  A cell changes due to a soft error caused for instance by an alpha 
        particle hit.  
    2.  A cell changes due to a voltage transient on logic ground.  In this 
        scenario the specific cells affected by such a transient are those
        that are 'weakest' at the time.   While many cells might be changed
        you will only know about the first one that a read attempt is made
        on (i.e., the one that generates the parity panic).
Note that in the two cases above the location of failure will probably
be random.

The design is extremely robust in handling spikes or large transients
across AC neutral and AC phase, however, in order to pass VDE, class B, 
et. al. the designed separated AC neutral, Safety Ground and Logic Ground.  
In normal user ac power situations this is no problem.  However, when the
user has problems such as floating grounds, or peripherals on one phase
and the computer on another phase (or whatever) a measurement of the
rms voltage between AC neutral and Safety ground will indicate the problem.
The Model 370 will NOT tolerate voltage greater than 1 volt rms between
these two lines.  Often a power line monitor is required in order to
catch transients across these two lines which sometimes occur as the
result of an external event (elevator motor, or ..).

To put all of this into perspective.  There are a lot of Model 370's out
there (in the tens of thousands).  You can count the sites that have 
experienced recurring parity problems on one hand.  In every previous
case we have found that curing input power problems solved the parity 
problems.

The normal comments about this not bein an official postion of H-P etc.
apply.

Glen Robinson