Re: On caches and parallelism
> From: "Corliss, George" <george.corliss@xxxxxxxxxxxxx>
> To: Nate Hayes <nh@xxxxxxxxxxxxxxxxx>, Dan Zuras Intervals
> <intervals08@xxxxxxxxxxxxxx>
> CC: "Corliss, George" <george.corliss@xxxxxxxxxxxxx>,
> "<stds-1788@xxxxxxxxxxxxxxxxx>" <stds-1788@xxxxxxxxxxxxxxxxx>
> Subject: On caches and parallelism
> Date: Mon, 4 Jul 2011 15:15:52 +0000
>
> Nate and Dan,
>
> I REALLY appreciate each of you taking the time to provide descriptions and=
> speculations on the architecture and applications considerations of caches=
> and parallelism. By quirks of history, I teach my department's Hennessy a=
> nd Patterson Computer Architecture course. Although it is beyond my expert=
> ise, I do the best I can.
>
> May I use each of your messages in my course when we get to the memory hier=
> archy chapter?
I'd be honored.
Although I don't know how authoritative my comments
may be considered nor how you are going to cite them. :-)
>
> Most of my students go on to develop business applications, and it am alway=
> s trying to help them see when they might actually need to know this Henne=
> ssy and Patterson stuff. Your messages might help.
>
> Thank you.
>
> Dr. George F. Corliss
I don't know Hennessy personally although he is now the
president of my school. I guess starting a successful
company will do that for you.
But I can tell you from personal experience that Patterson
has the advantage for an academic of actually having made
some chips. Although the making of chips at a university
via the use of graduate students is a VERY different
process than what happens in industry. It took him years
& its a wonder he ever finished at all.
BTW, in Patterson's time there were no on-chip caches.
Partly because it was before the time when such a thing
would have been feasible. But mostly because one of his
design principles was to match on-chip speeds with the
r/w speed of the RAMs of that era.
Over time, as CPU speeds became substantially faster than
RAMs, eventually that principle had to go. But it is still
a good idea to think about the bandwidth hierarichy as
Cray did. The man was way ahead of his time both in speed
& simplicity.
Dan