distributed computing implementations

[Martin v. Löwis]

Robin Becker <robin at jessikat.fsnet.co.uk> writes:

> so how do master slave implementations handle thousands of slaves? If
> the master is really a 'master' and not a SETI type server it would seem
> that these resource limits might play a part. How do GRID systems work
> this?

In a compute-intensive application, it is sensible to close the TCP
connection after you have communicated the job. Keeping the connection
alive is important to avoid the TCP connection setup, however, that is
neglible if you spend several minutes or more in computation.

I don't actually know how grid computing protocols work, but I believe
they are not concerned about networking performance, as they use HTTP
and SOAP to communicate jobs.

Regards,
Martin