Re: more fun with proposals

Eric Salo (salo@mrjones.engr.sgi.com)
Thu, 11 Jul 1996 14:13:15 -0700

> Exactly. In chemistry apps this would correspond to some large ( >100MB in
> size) distributed matrix with processes accessing dynamically different
> subsections (around 1KB in size). There is no predefined static access
> pattern, and therefore mapping these small subsections into separate windows
> to avoid concurrent accesses is not practical. Also locking is expected to be
> an unnecessary and heavyweight operation that would kill performance.

Once again I don't understand. If the access pattern is not known, how do you
prevent simultaneous puts to the same region without a lock?

I should back up a bit...I don't feel passionately that the strict locking
semantics are the only possible solution. I really like the simplicity and
elegance of the model, but if there are compelling examples of codes that need
more flexibility then we should do our best to accomodate them. But I still
haven't seen many concrete examples (that I understand) which could not be
handled well by either the more restrictive lock semantics or a pure
get/put/barrier model.

-- 
Eric Salo         Silicon Graphics Inc.             "Do you know what the
(415)933-2998     2011 N. Shoreline Blvd, 8U-808     last Xon said, just
salo@sgi.com      Mountain View, CA   94043-1389     before he died?"