Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rochester!rutgers!gatech!hubcap!brisco
From: brisco@caip.rutgers.edu (Thomas Paul Brisco)
Newsgroups: comp.hypercube
Subject: Re: ncube memory
Message-ID: <372@hubcap.UUCP>
Date: Fri, 7-Aug-87 14:54:17 EDT
Article-I.D.: hubcap.372
Posted: Fri Aug  7 14:54:17 1987
Date-Received: Sun, 9-Aug-87 03:45:39 EDT
Sender: fpst@hubcap.UUCP
Lines: 25
Approved: hypercube@hubcap.clemson.edu

In article <370@hubcap.UUCP> Lonnie R. Welch (welch-l@osu-eddie.UUCP) writes:

>    I am trying to write a
> program that requires the use of several large data structures.  It seems
> that the Host process can only have a data segment of size <= 64k.  None of
> the variables in my program are larger than 64k, but the sum of the memory
> required by all the variables is > 64k.  Has anyone else found this to be a
> problem?  How did you get around this limitation?

	You don't state whether you acutally need all of the variables
at a given time or not, but one of my tricks is to manually "data
swap" -- i.e. keep data that is not immediately needed on disk, and
malloc()/free()  space as necesary to achieve you means.
	One other, more crafty but harder, way is to split your
program into 2 processes and do the operations seperately, but
communicating (results) via pipes -- I've found this to be handy when
there are 2 large computations are involved (where both must be on the
host processor), but only the results are really necessary to proceed
in the original process.  Look in n_frun, and see whether you really
need the computations locally, or whether they can be moved out of
(what I call) the "controlling process".

					Tp.
                  -                  ARPA: Brisco@rutgers                  -
                  -  UUCP: (ihnp4!ut-sally, allegra!packard) !caip!brisco  -