750 dies in fsck

[Jeffrey Jongeward]

A local startup company has purchased a 750 from an OEM
and 4.1 bsd binary license from a reputable Unix s/w house (whom
I am inclined to think has provided good s/w).
The 750 has twin eagles + a emulex sc750 controller.

However, when dma-ing the same large block ( >8K ) from a raw file 
system (e.g. rhp0g) into memory multiple times you find that 
randomly (i.e. on about 1 out of 5 reads) there is disagreement over 
what is read - typically 8-20 bytes within the buffer will disagree.  
There seems to be a pattern to this disagreement - Not always the 
same bytes within one of the buffers, but often certain bytes offsets 
reoccur quite often.  This causes havoc with fsck and with paging.

sc750's have been changed, both drives have been used, memory
controllers have been changed, memory exchanged, etc., but to
no avail.  The OEM is short on boards and when attempting to
swap the cpu, they (somehow) blew the memory controller, so
a cpu swap has not been tried.

Has anyone else experienced this problem?  Is there some
incompatibility between certain 750 board rev. levels 
and certain sc750's?. Is this a timing problem, etc., etc.  
Does anyone know / has anyone experienced this before?


			This little startup company
			would thank you in advance
			had they a 750 that could
			be on the net,

			uw-beaver!ssc-vax!jeff