From:	MERC::"uunet!CRVAX.SRI.COM!RELAY-INFO-VAX" 16-JAN-1993 23:30:47.10
To:	info-vax@kl.sri.com
CC:	
Subj:	Re: Long standing apparent Backup or XQP problem

In article <1993Jan7.231426.1@cc.curtin.edu.au>, zrepachol@cc.curtin.edu.au writes:
> In article <9301052330.AA21443@relay1.UU.NET>, gleeve@galaxy.dnet writes:
...

A foot note to my previous posting...

We also run PhaseII vol shad. Turning it off removes the errors... :-{

Many hours of comptemplation have got me to think that this is also a load
and/or timeing thing. My theory is that backup get the IO completion BEFORE
all the data is in the buffer, charges down the bytes computing CRC on new
AND OLD data. By the time the QIO to output the data happens, all the data
is where it should be, but the CRC is off due to the stale data in the buffer.

If this is the case, it is ***REAL*** nasty, as; no warning or error message.
If the 'old' and 'new' data are highly alike, and regulaly formated, the
incorrect values may pass all but the most detailed checking ( back/list is
the only one that would do this I think, maybe decnet as well?? ). The high
sencetivity to load, configuration etc means it is very unstable, possibly
becoming a major problem in a future VMS version or on a new CPU. It seems
to be in a very critical section of IO code.

~Paul