3L Parallel Fortran : Processor Farms

Thread View: comp.sys.transputer

2 messages

[T] Back to Threads [V] Tree View [G] Group Articles

2 total messages Started by Dzaharudin Mansor Mon, 08 Oct 1990 02:10

#1023

Author: Dzaharudin Mansor
Date: Mon, 08 Oct 1990 02:10

40 lines
1672 bytes

Hi,
	I'm developing a application program using the 3L
	Parallel Fortran compiler. The application is
	of the type that requires a master, and as many 
	worker tasks as possible.....hence the use of the
	Processor Farms application model using the flood-fill
	configurer.

	The problem is that I am not getting good speedup
	from 1 Transputer to 4 Transputers (only a 2X speedup
	when using 4 Transputers), and on examining 
	how the work packets are sent and result packets received,
	It seems that there are only two worker tasks in the network.

	Possible explanations:
		1. The configurer sets up only two worker tasks
		   for the 4 Transputer configuration.

		2. The master processes are scheduled in a way that
		   the use of the external worker tasks is not optimal.

	Anyone else had similar experiences?

	I would also like to seek information on:

	1. How does the flood-fill configurer determine how many
	   worker tasks that will be employed given X
	   number of Transputers.

	2. What type of scheduling does the compiler employ?
	   Time slicing? Preempt on waiting for communication?



	Any views, references, discussion (anything at all) will
	be greatly appreciated.

	Thanks.

-dzahar-

re: 3L Parallel Fortran

#1027

Author: coms0303
Date: Mon, 08 Oct 1990 20:05

37 lines
1849 bytes

In article <8922@latcs1.oz.au>, dzahar@latcs1.oz.au (Dzaharudin Mansor) writes:
> 	The problem is that I am not getting good speedup
> 	from 1 Transputer to 4 Transputers (only a 2X speedup
> 	when using 4 Transputers), and on examining 
> 	how the work packets are sent and result packets received,
> 	It seems that there are only two worker tasks in the network.
> 
> 	Possible explanations:
> 		1. The configurer sets up only two worker tasks
> 		   for the 4 Transputer configuration.
> 
> 		2. The master processes are scheduled in a way that
> 		   the use of the external worker tasks is not optimal.
> 
> 
> 	Any views, references, discussion (anything at all) will
> 	be greatly appreciated.
> 
> 	Thanks.
> 
> -dzahar-
We had similar problems with the 3L flood fill configurer, and never managed
to get satisfactory results from it. In the end we wrote our our version
of the configurer that will load a set of identical processes on to any
array of transputers, and a 3L C multiplexor program to connect this to
the host.

3L are very helpful, and have a technical note on configuration. From there
it's just a matter of looking at some files generated by the configurer to
understand the commands that the 3l loader (gloader.b4) uses, and to generate
your own.

Ian Graham, Computer Science, University of Waikato, NZ

PS. If anyone out there is looking for a lecturer/senior lecturer job with
possiblities for transputer research - we have several vacancies.

Thread Navigation

This is a paginated view of messages in the thread with full content displayed inline.

Messages are displayed in chronological order, with the original post highlighted in green.

Use pagination controls to navigate through all messages in large threads.

Back to All Threads