Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!pur-ee!hankd
From: hankd@pur-ee.UUCP (Hank Dietz)
Newsgroups: comp.arch
Subject: Re: flexible caches
Message-ID: <12907@pur-ee.UUCP>
Date: 18 Sep 89 19:51:38 GMT
References: <224@qusunr.queensu.CA> <22151@cup.portal.com> <2115@munnari.oz.au> <22211@cup.portal.com>
Reply-To: hankd@pur-ee.UUCP (Hank Dietz)
Organization: Purdue University Engineering Computer Network
Lines: 23

In article <22211@cup.portal.com> mmm@cup.portal.com (Mark Robert Thorson) writes:
[...as to use of neural nets to dynamically adjust caching...]
>The speed of learning depends on the model being used.  Would you have
>agreed with my statement if I had used the term "advanced statistical
>techniques" rather than the emotion-laden "neural networks"? ....

No, I would not agree.  I do prefer calling this "statistical prediction" --
because that's what it is ;-).  Many optimizing-compiler writers have
considered improving the compiler's analysis by using stats collected by
trial executions, but it rarely helps.  Statistics based on runtime data are
notoriously unstable.

Remember that, given the (stochastic) reference pattern, compiler algorithms
do exist for determining optimal cache control...  see Chi's PhD thesis.

>Also, note that many programs provide the opportunity to gather statistics
>over thousands of runs.  A display list interpreter for computer animation
>might get called 60 times a second.  1000 runs would take less than 17 seconds.

An interesting example...  one doesn't usually recompile for each display list.
I'll get back to you on this one....

							-hankd@ecn.purdue.edu