Re: [9fans] Parallel computing and Plan9

Re: [9fans] Parallel computing and Plan9

[dpx]

[-- Attachment #1: Type: text/plain, Size: 274 bytes --]

One idea for message passing is to use pipes. Plan9 make it easy
to start pipes and then import them from remote machines. In our
group here we are in the early stages of trying this out. There
is some very preliminary info at http://www.acl.lanl.gov/plan9/mp/

-dp


[-- Attachment #2: Type: message/rfc822, Size: 3106 bytes --]

From: Ish Rattan <ishwar@pali.cps.cmich.edu>
To: <9fans@cse.psu.edu>
Subject: [9fans] Parallel computing and Plan9
Date: Fri, 20 Jul 2001 15:17:10 -0400 (EDT)
Message-ID: <Pine.LNX.4.30.0107201516300.1586-100000@pali.cps.cmich.edu>

Hello:

I have been thinking about parallel programming under Plan9.

Traditional way to write parallel program is to partition the task
into a number of subtasks and control their execution-scheduling and
communication among a number of autonomous systems.

In Plan9 case, the above scenario does not hold. The computations are
done on a cpu-server. If there is only one cpu-server then the parallel
programming is no different form non-parallel case (except for
concurrency control). Say there are two cpu-servers in a system:
1) can one distribute the computation on these cpu-servers?
2) what about the interaction between the distributed computation parts?

I did not find any technique for a program to discover the cpu-servers
in the system too.

Multithreading/light-weight processes in the same computation do help
but is not strictly speaking 'shared memory' or 'message passing'
communication based parallel programming. So how does Plan9 fit in
the parallel programming paradigm that uses cpu cycles from different
cpus to reduce the computation time?

I would like your comments, as I plan to pursue/explore this issue Fall
semester.

-ishwar

[9fans] Parallel computing and Plan9

[Ish Rattan]

Hello:

I have been thinking about parallel programming under Plan9.

Traditional way to write parallel program is to partition the task
into a number of subtasks and control their execution-scheduling and
communication among a number of autonomous systems.

In Plan9 case, the above scenario does not hold. The computations are
done on a cpu-server. If there is only one cpu-server then the parallel
programming is no different form non-parallel case (except for
concurrency control). Say there are two cpu-servers in a system:
1) can one distribute the computation on these cpu-servers?
2) what about the interaction between the distributed computation parts?

I did not find any technique for a program to discover the cpu-servers
in the system too.

Multithreading/light-weight processes in the same computation do help
but is not strictly speaking 'shared memory' or 'message passing'
communication based parallel programming. So how does Plan9 fit in
the parallel programming paradigm that uses cpu cycles from different
cpus to reduce the computation time?

I would like your comments, as I plan to pursue/explore this issue Fall
semester.

-ishwar