From: Thilo Kielmann Newsgroups: comp.parallel.mpi Subject: Re: timing for Wait(all|any) Date: Thu, 8 Jul 1999 08:13:25 GMT Organization: Fac. Wiskunde & Informatica, VU, Amsterdam Message-Id: <7m1mj5$eed@cs.vu.nl> References: User-Agent: tin/pre-1.4-19990413 ("Endemoniada") (UNIX) (SunOS/5.5.1 (sun4u)) Cancel-Lock: sha1:UQDwF9Hqou1JDB6iQqDFs3lusA0= Xref: ukc comp.parallel.mpi:5301 Johannes Zellner wrote: : recently I tryed a one-master - multiple-servers program. : The master sends jobs to the servers (Send) and sets up (Irecv)'s : for each server. Then it goes to a (Waitany) and if this returns : does another (Send)/(Irecv) for the particular Waitany's Request.source(). : As the server jobs take quite a while compared to the master-server : traffic, I supposed my master to eat almost no cpu. : But: running this on a single machine with for example `-np 3' : and each process gets about 33% cpu, also the master. : what's going on there? (How is the Wait stuff implemented) As you do not write which operating system/MPI implementation you are using, I strongly doubt that somebody will come up with a hint how Wait is implemented on your platform. Anyway, if your processes exchange a whole lot of data (e.g. megabytes) then it may well impact your performance, esp. when many workers simultaneously try to send to the master. Then, a better overlapping of computation and message sending will help you. You may wish to have a look into: Bernd Freisleben, Dieter Hartmann, and Thilo Kielmann. Parallel Raytracing: A Case Study on Partitioning and Scheduling on Workstation Clusters. Proc. Hawai`i International Conference on System Sciences (HICSS-30), volume 1, pages 596-605, Wailea, Hawai`i, USA, January 1997. http://www.cs.vu.nl/~kielmann/papers/hicss97-sched.ps.gz Good luck, Thilo -- Dr. Thilo Kielmann Dept. of Mathematics and Computer Science, Vrije Universiteit De Boelelaan 1081a, 1081 HV Amsterdam, The Netherlands Phone: +31 20 44 47789 /\,/\ Fax: +31 20 44 47653 / u u \ E-Mail: Thilo.Kielmann@acm.org \<_Y_>/ WWW: http://www.cs.vu.nl/~kielmann/ / " \