[Spock@SAMSON.cadr.dialnet.symbolics.com: IFU]

[Jerry Bakin <Jerry@marlowe.inference.dialnet.symbolics.com>]

    Date: Thu, 1 Dec 88 18:50 EST
    From: barmar@Think.COM (Barry Margolin)

    We've had LOTS of problems getting reliable IFUs and XSQs.  We started
    having trouble with one of our 3645s this summer (generally, crashes
    were the only symptom).  We replaced both boards several times, and
    continued to have problems.  We downgraded the machine to a 3640 while
    our CE tried to find us a working board set.  He kept ordering new sets,
    and he would run them in his test system for several days.  It took him
    over a month to get a set that worked reliably.

    A common problem with these, according to our CE, is getting a set of
    boards that work well together.  At the parts depot they do some testing
    of the boards, but on an individual board basis, not as whole sets.
    Sometimes different boards have slightly different tolerances, so an IFU
    that worked in the depot's test system might not work in your system.

    Our CE has also complained that the DOA rate of IFUs is around 75%.

    Unfortunately, the performance benefits of a working IFU seem to justify
    the difficulty of getting one.

						    barmar

My 3645 also had severe problems this summer, and it took six weeks to
get it working again -- I'm knocking on wood and hoping it doesn't
understand what I type in....

I have seen this problem on at least five different 3645s for over two
years, and I have yet to see Symbolics address it as a generic 3645
problem.  Each time it occurs the CE of the week scratches his head and
starts swapping boards.

I could live knowing the problem existed, if Symbolics had some way of
fixing it on a timely basis.  Right now when the techs start swapping
boards they turn me into an involuntary Symbolics QA observer.  Not
good: not until Symbolics starts paying me and reimbursing my company
for lost time.

I don't know how frequently the problem occurs, perhaps Symbolics has
some statistics on how often 3645 boards are replaced in comparison to
3640s.  I would suggest something as dramatic as Symbolics maintaining
two or three 3645 board sets in a hot and running state and when someone
reports the problem, just swapping the board sets completely.
Presumably they can then examine/identify the real culprit as it occurs
in their lab on the old board set, and in the mean time, I will not have
to waste six weeks with a completely unreliable lispm while perusing the
ads in PC World, Byte, and MacWorld.

On the otherhand, I'm not sure what finally made the problem disappear,
but the last thing our CE did was to adjust the power supply.

Jerry.