RA81 disk curdling

[haynes%ucsce.ucscc.UUCP%Berkeley at ucscc.UUCP]

We have had some episodes of this, evidenced by no hardware error messages,
but fsck finding partially allocated inodes, doubly allocated blocks,
files trashed, etc.  Altho the UDA50 and RA81 are supposed to be
quite thoroughly self-testing internally, the Unibus isn't.  Over many
years we have had troubles with noise sensitivity on the Unibus causing
missing or extra words to be transferred to DMA devices, which can
really make a mess of your disk without being detected by the hardware.
Nor is it usually detected by diagnostics, as it requires lots of bus
activity to make it happen.

In our recent case we had a DQS11 sitting on the Unibus not being used.
It appears the disk corruption has stopped now that we have removed
the DQS11 from the bus entirely.  So there is probably a bus driver or
receiver in the DQS11 that is dragging some Unibus line down near its
threshold. (Since the DQS11 did work without disk corruption at one
time, it is something that has deteriorated over time.)

DEC has, or used to have, a tester that might show this kind of problem.
It was a bus terminator connected to a variable-voltage power supply,
so the bus signal high level could be run up or down from its nominal
value.

We're never really going to be rid of this problem until the Unibus is
replaced by something that at least has parity.  It would help if we
just had all the DMA devices like disks off the Unibus, leaving it for
things like DZ11s where an occasional missing or extra character might
not bother anybody much.

ucbvax!ucscc!haynes