[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: MCL size limitations (was "Can I access 256MBytes of memory?")
- To: David Kieras <firstname.lastname@example.org>
- Subject: Re: MCL size limitations (was "Can I access 256MBytes of memory?")
- From: email@example.com (Bill St. Clair)
- Date: Fri, 5 Feb 1993 16:25:51 -0600
- Cc: info-mcl, dambrosi@chert.CS.ORST.EDU
>Does the proposed 4mb vector limit apply to one dimensional arrays as well?
>That definitely won't work for me! I regularly allocate arrays larger than
>that in computing expected utilities. (Application is real-time diagnosis and
>repair, using decision theory - large matricies of conditional probabilities).
>From: David Kieras <firstname.lastname@example.org>
>One of my major applications, involving natural language processing, uses
>a lexicon in which words are lisp symbols; it is easy to get thousands
>of symbols defined this way ... as I read your message about vector size,
>I could get into trouble if reducing the maximum vector size will cause
>substantial performance degradation when there are a lot of symbols.
You didn't quite read what I intended to communicate. There is a bug
in the current MCL that makes INTERN & FIND-SYMBOL slow down significantly
if a package contains more that 16000 symbols. "package-hash-patch", which
will be part of patch 3 for MCL 2.0, fixes that bug.
It is possible that MCL 2.1 will ship with a reduced maximum vector
size. This means that the maximum size of a general array will be
1 million elements and the maximum number of keys in a hash table
will be 500,000 elements. In MCL 2.0 as it currently stands, the
maximum array size is 4 million elements and the maximum number of
elements in a hash table is about 2 million.
I think that reducing the maximum vector size is a bad idea, but since
increasing the maximum back to MCL 2.0's level will require some effort
on my part, I asked for feedback. I would like to increase the maximum
array size so that it is limited only by the amount of memory that
the Macintosh can address, but doing this will require even more effort,
so I hesitate to do it unless lots of people really want it.