From: Tom Goodale Newsgroups: comp.parallel.pvm Subject: Re: What good is it, anyway!? Date: Wed, 21 Jul 1999 18:59:44 +0200 Organization: Max-Planck-Institut fuer Gravitationsphysik Message-Id: <3795FC80.919269AA@aei-potsdam.mpg.de> References: <7n22d0$u14$1@nnrp1.deja.com> <37964C07.846071F6@it.teithe.gr> <3795DC15.DE9A2340@Mutz.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Xref: ukc comp.parallel.pvm:8594 Marc Mutz wrote: > > Vasileiadis Spyros wrote: > > > > Well i fell the same way my friend so I decided to go one step ahead (if > > this is possible) I am now trying to figure out a way toy hack linux > > kernel so that new created threads will be distributed to the clusters > > without effecting the behavior. So every existing program that simply > > forks will run in parallel without reprogramming or recompiling!!!!! > > The programmer will just think that he is working on shared memory > > multiprocessor system and not on a distributed memory parallel > > clustering system > > > > any syggestions are wellcomed > > > Have you studied the linux kernel well enough? I think you will run into > major problems when trying just to 'hack' the linux kernel. You are > going to change the most fundamental routines of linux, like the > scheduler, shm, etc. > But this will not suffice. How will you tell which /dev/sda1 mount wants > to mount if it may run on any machine in the cluster? > How will you tell whether the fork you just had will only be there to > wait for blocking I/O and then exit (in that case transferring the whole > vm space of the process over the network just to wait for a few kB of > data would be a real performance eater :-) or do lengthy calculations on > a small amount of data (in that case it would be good wrt. performance)? > > Suggestions? Only one: Before throwing away your time, subscribe > yourself to the linux-kernel mailinglist and discuss this. Then try to > find people that will support you. This _much_ work! > > Marc > > -- > Marc Mutz http://marc.mutz.com/ > University of Bielefeld, Dep. of Mathematics / Dep. of Physics > > PGP-keyID's: 0xd46ce9ab (RSA), 0x7ae55b9e (DSS/DH) You could also take a look at MOSIX http://www.mosix.cs.huji.ac.il/txt_main.html which might fill your needs. Tom