Re: IFU on 3600

[MILLER@VAX.CAM.NBS.GOV]

In response to discussion about 3600's with IFU.
	We've got an older 3600 which we upgraded with the IFU around
Rel 6.1 time.  After the initial period of board swapping, it worked
fine and the performance gain was, all in all, worthwhile (although it
wasn't quite the 50% implied). 
	However, after Genera was installed, we started getting
mysterious bugs similar to what you are experiencing: After a number of
hours of hard symbol (and number) crunching, the machine would go into
the debugger. It was always the same error; a NIL ended up in the
arglist where a number had been in the internal function
SI:ARITHMETIC-FUNCTION-PLUS-BIGNUM-BIGNUM. I really can sympathize with
the `Frustration factor': I came in many a morning expecting to see
wonderful results from an overnight run and !)@#*)!@*#.  The bugs went
away when I rebooted Rel 6.1 (ugh).
	Happy ending: Nobody seemed to have a clue at Symbolics or
elsewhere what it might be, but Dave Plummer came up with a suggestion,
bless his heart. It seems that swapping boards randomly missed the
point. There existed, at that time, a magical list of `compatible' rev
levels for various boards.  With CPU board ver. X and Foo board ver. Y,
then you should use IFU board ver U, and XSQ board V. In hindsight it
seemed obvious, but in foresight it apparently wasn't. In any case, the
machine has been working like a charm ever since. Except the console is
flaky again, but thats another story....
			Bruce Miller
			MILLER@VAX.CAM.NBS.GOV