Newsgroups: comp.parallel.pvm
From: fender@olympus.crs4.it (Claudio Gallo)
Subject: pvm timing
Keywords: timing
Organization: CRS4, Center for Adv. Studies, Research and Development in Sardinia
Date: 6 Aug 1994 10:58:08 GMT
Message-ID: <31vqc0$l2s@berlioz.crs4.it>

In article <31u9rh$j1p@blaze.rutgers.edu>, jjiao@blaze.rutgers.edu (Jia Jiao) writes:
|> 
|>     How do you people time the pvm programs? I use "gettimeofday" but
|> it is very unstable, depends on the processor load, network traffic etc...
|> On the other hand the Fortran timer etime seems to give consistent timing
|> but it seems to exclude the node time. (Is it CPU time?)
|> 
|>    Thanks.
|> 
|>   Please respond to jjiao@cs.rutgers.edu

I would not say that etime SEEMS to exclude the node time, but IT REALLY DOES IT!
Infact functions of that sort (etime for DEC, mclock for AIX, etc. ) measure 
(this is something I say from my personal experience) 
only the CPU time used for computations, but when the 
program get into the "sleep" phase caused by the unix-priority-queue scheduling,
etime goes to sleep 'till the moment in which it'll get back to "run" phase.
When a master casts the informations to the slaves it get into to sleep-phase
until it'll not have a call back from the slaves (at least in the case in which it has not
something else to do), so you loose communication's and the slave's work-time.
If I were in you I would use both the measure, and I would use "gettimeofday" on the slave
too (dumping the data in some files whose name include the slave PID). 
With some simple subtractions you should be able to get a better timing.
Anyway I would suggest to try to rule things in order to impose the machine you are
logged on to be the console. Your measures should be more reliable because they 
should not be affected by the remote time sinchronization (this is something you
can choose in PVM 3.3.X). 
Furthermore I would like to point out that it is not easy  to obtain a good timing
using gettimeofday whose precision is of more or less of the order of magnitude of
the second ...

Anyway, I hope that I have given you some useful hints...

Cheers!

-- 

   -\   /-          '||''''|                      '||
    @   @            ||  .     ....  .. ...     .. ||    ....  ... ..
     . .             ||''|   .|...||  ||  ||  .'  '||  .|...||  ||' ''
 .         .         ||      ||       ||  ||  |.   ||  ||       ||
 |\_______/|        .||.      '|...' .||. ||. '|..'||.  '|...' .||.
 | \/   \/ | 
  \_______/         '''''''' ''''''' '''''''' '''''''' ''''''' '''''''
                   "Mantua me genuit, Calabri rapuere;
fender@crs4.it      tenet nunc Partenope; cecini pasqua, rura, duces ..."


