[Beowulf] Definition of HPC

[Greg Keller]

>
> From: Adam DeConinck <ajdecon at ajdecon.org>
> Subject: Re: [Beowulf] Definition of HPC
> To: beowulf at beowulf.org



My own (vague) definition of HPC would probably be based more on the
> intended design of the system, something like:
>
> "A high-performance computing system is designed to use one or more
> computers to achieve the maximum performance possible on a particular
> computational task."
>
> This could be improved, but it makes a distinction between maxing out a
> "normal" computer and designing a system specifically for performance.
>

The definition I use is:
A Computational task that requires combining the performance of multiple
systems to achieve results faster than a single system.  The key
requirement is the aggregation of hardware resources (CPU, Memory, Disk,
Networking, etc) beyond a single "box" whether under a single OS instance
like ScaleMP/SMP or multiple OS/s using MPI and other methods of managing
the distribution of tasks.

This should be easily distinguishable from Enterprise computing tasks which
often can make due with a fraction of a system to complete its designated
task, enabling the virtualization technologies to thrive.  Multiple systems
are used typically for High Availability rather than High Performance in
Enterprise computing.

The "task" still needs to be defined, perhaps as a "computational task", to
distinguish it from farms of email servers that may well churn at 100%
utilization at peak times.  Of course, email looks more and more like a big
data / big compute problem every day when hiding spam and virus protection
tasks are taken into account.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20130415/239f5150/attachment.html>