From: Paul Wilson
Subject: Re: Hardware Caching and LISP applications, any experience?
Date:
Message-ID: <m6vjcjINNe6c@jive.cs.utexas.edu>
In article <··········@netnews.jhuapl.edu> ·······@aplcomm.jhuapl.edu (Robert Evans) writes:
>I am in the process of upgrading some Sun workstations whose primary
>purpose is to run Lisp applications. I have heard some rumors that
>the 1MB cache found on the newer Sparc10's has actually slowed down
>performance in the LISP environment due to the excessive memory size
>and cache misses found in the LISP world.
I suspect that the larger cache is going to be good for you IF your
youngest generation (of your presumably generational garbage collector)
fits inside and pretty much stays there. (I've done some research on
this---if you're interested, see the paper cache.ps, ftpable per the
instructions in my .signature.)
If the youngest generation doesn't fit in the cache, you get cache
misses proportional to (and slightly greater than) the rate of allocation.
If your cache miss service is slow, that's bad news, so conceivably
a larger cache could cause a slowdown---you'd get about the same number
of misses, but they'd cost more. A cache of 1 MB should be able to
hold a reasonable-sized youngest generation, though, so that shouldn't
be a problem if your GC is configured right. (This may not be true
if you're running a bunch of Lisp processes, instead of only one, or
are competing with other processes with a big cache footprint.)
>Does anyone have any experience and/or performance data in this area.
>Specifcally, does a Sparc 10/41 run lisp better or worse than a Sparc
>10/40?
Please post any such info if you've got it.
-- Paul
--
| Paul R. Wilson, Computer Sciences Dept., University of Texas at Austin |
| Taylor Hall 2.124, Austin, TX 78712-1188 ······@cs.utexas.edu |
| (Recent papers on garbage collection, memory hierarchies, and persistence |
| are available via anonymous ftp from cs.utexas.edu, in pub/garbage.) |