octave-maintainers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 2nd: sparse matrices


From: Andy Adler
Subject: Re: 2nd: sparse matrices
Date: Tue, 27 Apr 2004 14:37:40 -0400 (EDT)

On Mon, 26 Apr 2004, Etienne Grossmann wrote:
> I have been atracted to this thread because I myself have sometimes
> slowness probs w/ matrices w/ ~5e7 doubles.
>
>   My ignoravimus' question is : doesn't the OS (Linux) use its swap
> disk space to allocate space beyond the memory capacity of the
> computer it is running on? Of course, it will swap data back and
> forth, not necessarily as efficiently as if I moved the data myself,
> but isn't the idea of swap that it should be done by the OS. Iirc, VMS
> had a reputation for good virtual memory management.

OS use of swap space is optimized in a very generic way. It is
possible to do far, far better with knowledge of the algorithm.

>   About the sparse package in general, its point is, iigc, to avoid
> both storing and adding/multiplying/whatever lots of zeros. As the
> name 'sparse' says, it is for matrices w/ few nonzeros, not
> necessarily 'huge'.

There do exist all sorts of specialized libraries:
   - out of core solvers
   - distributed solvers,
     etc.

It would be interesting for someone to look at these and
think about how to integrate with octave.

However, the current (Matlab inspired) approach is IMHO
quite well thought out and solves a certain set of
problems well.

If someone would like to propose and design an interface
to these specialized libraries, I would be an enthusiastic
cheerleader.

Andy





reply via email to

[Prev in Thread] Current Thread [Next in Thread]