Newsgroups: comp.parallel.pvm
From: R.Abdullah@lut.ac.uk (R Abdullah) 
Subject: Questions from a new PVM user
Organization: Loughborough University, UK.
Date: Wed, 1 Jun 1994 17:26:47 GMT
Message-ID: <CqqB4n.65y@lut.ac.uk>

Hello there!

I'm working on my first PVM program and have the following questions to ask.

1. My program seems to work fine when I use 1 or 2 slaves. When I start using >2 slaves, it sometimes work and sometimes doesn't. When it doesn't work, I halt
PVM, restart it again and run the program again. Sometimes it works and sometimes it does not work again. I find this very peculiar. 
Can someone explain this peculiarity? 

2. There is not much of a difference in the time taken to execute between using 1 slave and 2 slaves, even on a 300 by 300 matrix. The same algorithm when run on a shared memory machine showed reduction in time by about one half. Apart from communication time, what else could contribute to this?

3. Intuitively, I would time my program by taking the start time to be before 
the master sends data to the slaves. The final time is when the slave has 
received all results from the slaves. 
Alternatively, the start time could be before each slave begins computation and the final time is when the slave has finished computation. In this case, the time is that of the slowest slave. 
Which of these (or there might be other methods) is normally used and why?

Please respond my mailing me directly at R.Abdullah@lut.ac.uk

I would also appreciate if anyone could suggest reading materials realted to the above questions.

Thank you very much.



