Menu
▾
▴
bproc-users — General discussion about BProc.
You can subscribe to this list here.
| 2001 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
(25) |
Nov
|
Dec
(22) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2002 |
Jan
(13) |
Feb
(22) |
Mar
(39) |
Apr
(10) |
May
(26) |
Jun
(23) |
Jul
(38) |
Aug
(20) |
Sep
(27) |
Oct
(76) |
Nov
(32) |
Dec
(11) |
| 2003 |
Jan
(8) |
Feb
(23) |
Mar
(12) |
Apr
(39) |
May
(1) |
Jun
(48) |
Jul
(35) |
Aug
(15) |
Sep
(60) |
Oct
(27) |
Nov
(9) |
Dec
(32) |
| 2004 |
Jan
(8) |
Feb
(16) |
Mar
(40) |
Apr
(25) |
May
(12) |
Jun
(33) |
Jul
(49) |
Aug
(39) |
Sep
(26) |
Oct
(47) |
Nov
(26) |
Dec
(36) |
| 2005 |
Jan
(29) |
Feb
(15) |
Mar
(22) |
Apr
(1) |
May
(8) |
Jun
(32) |
Jul
(11) |
Aug
(17) |
Sep
(9) |
Oct
(7) |
Nov
(15) |
Dec
|
|
From: Erik A. H. <er...@he...> - 2001-12-04 23:35:38
|
On Mon, Dec 03, 2001 at 12:55:52PM -0500, henken wrote: > The attached oops occured on bproc 3.1.1 using a bproc-patched stock > redhat 2.4.9-6. The test script that I have been running that triggers > this is: I haven't been able to reproduce it on the systems here. I don't run Red Hat kernels on any of the systems here though. I suspect this is not a BProc problem. .looking at the back trace, it looks like it died in a call from somewhere inside the IP code. aic7xxx is mentioned in there too which seems odd but I don't really understand how to read those lines in your back trace since they also mention bproc. Has anyone else seen anything like this? - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: henken <he...@se...> - 2001-12-03 17:55:57
|
The attached oops occured on bproc 3.1.1 using a bproc-patched stock
redhat 2.4.9-6. The test script that I have been running that triggers
this is:
#!/bin/bash
JOBID=${0##/*/}
/bin/echo "JOBID:$JOBID"
NODES=`/usr/local/clubmask/bin/getnodes $JOBID`
/bin/echo "NODES:$NODES"
for node in $NODES; do
( let count=0
while [ $count -le $1 ]; do
/bin/echo "Number: $count on node: $node"
bpsh $node /home/henken/e2k/jobs/nic
let count=count+1
done ) &
done
==========================================
the binary program nic is:
#include <unistd.h>
#include <stdio.h>
int main() {
int pid;
pid = fork();
if(pid) printf("Hello!!!\n");
else printf("Goodbye\n");
}
=========================================
I have been running this on anywhere from 2 to 8 nodes with $1 ( the
number of iterations ) somewhere over 100,000 to 1,000,000
Nic
--
Nicholas Henke
Undergraduate - SEAS '02
Liniac Project - University of Pennsylvania
http://clubmask.sourceforge.net
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
There's nothing like good food, good beer, and a bad girl.
|
|
From: J . A . M. <jam...@ab...> - 2001-10-25 22:23:54
|
Hi. Have you seen this in kernel list: http://marc.theaimsgroup.com/?l=linux-kernel&m=100403076708948&w=2 ??? Could this supercede ksyscall module ? If so, it will be good that you check that all the calls needed by bproc are exported... -- J.A. Magallon # Let the source be with you... mailto:jam...@ab... Mandrake Linux release 8.2 (Cooker) for i586 Linux werewolf 2.4.14-pre1-beo #1 SMP Thu Oct 25 16:19:19 CEST 2001 i686 |
|
From: Erik A. H. <er...@he...> - 2001-10-22 20:22:52
|
On Mon, Oct 22, 2001 at 09:55:53AM -0700, Shane Walton wrote: > On an initial connection of any slave to the master, > the node has the state of "error" which doesn't seem > to matter and defaults to ownership of "root:any". Is > there any way to make a slave connect with a desired > default state such as "up" and "any:any"? Thanks. BProc 2.2.x does this if the node setup script (/etc/beowulf/node_up) is missing or exits with non-zero status. If that script exits with zero status (an empty script is ok), then it should default to up/any/any. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: Erik A. H. <er...@he...> - 2001-10-22 20:21:15
|
On Mon, Oct 22, 2001 at 02:04:33AM +0200, J . A . Magallon wrote: > And how do I make bproc (or MPI) know there are two processors on the box > (ie, two possible slave processes of the same master could run there) ? > Suppose I add one other box with just 1 cpu...I should have 11 slaves. > I know that part of the question is MPI or PVM specific (I remember somthing > like weights in PVM, do not know about MPI). But can this be done > at the bproc level ? I believe you can simply list appropriate hosts twice to get MPI to run multiple jobs on a single host. BProc is not a scheduler or a message passing library so it doesn't handle any aspect of this. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: J . A . M. <jam...@ab...> - 2001-10-22 00:04:42
|
On 20011022 Erik Arjan Hendriks wrote: >On Mon, Oct 22, 2001 at 01:47:28AM +0200, J . A . Magallon wrote: >> Hi. >> >> I am now in the process of building a beowulf system with dual SMP boxes. >> System is made with a front end and five dual boxes. >> >> And the question is...I see two policies to follow: >> - Run one bpslave on each node, so system looks lie a 5-cpu one, and you could >> run two jobs in parallel with no overhead > >I strongly recommend this arrangement. Running a second slave daemon >is just added overhead. Plus, having separate slaves makes it more >difficult for things like MPI to know that they can take advantage of >shared memory, etc. > And how do I make bproc (or MPI) know there are two processors on the box (ie, two possible slave processes of the same master could run there) ? Suppose I add one other box with just 1 cpu...I should have 11 slaves. I know that part of the question is MPI or PVM specific (I remember somthing like weights in PVM, do not know about MPI). But can this be done at the bproc level ? -- J.A. Magallon # Let the source be with you... mailto:jam...@ab... Mandrake Linux release 8.2 (Cooker) for i586 Linux werewolf 2.4.12-ac3-beo #4 SMP Thu Oct 18 21:52:06 CEST 2001 i686 |
|
From: Erik A. H. <er...@he...> - 2001-10-21 23:56:31
|
On Mon, Oct 22, 2001 at 01:47:28AM +0200, J . A . Magallon wrote: > Hi. > > I am now in the process of building a beowulf system with dual SMP boxes. > System is made with a front end and five dual boxes. > > And the question is...I see two policies to follow: > - Run one bpslave on each node, so system looks lie a 5-cpu one, and you could > run two jobs in parallel with no overhead I strongly recommend this arrangement. Running a second slave daemon is just added overhead. Plus, having separate slaves makes it more difficult for things like MPI to know that they can take advantage of shared memory, etc. > - Run two slaves on each box, to have the system looke like a 10-cpu box. > > How can the second setup be configured ? The main problem doing that is that the master expects only 1 IP per slave. You can get arround that by setting up an IP alias on the slave for the second bpslave. Then use the '-s' option to specify a source address for the slave connection. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: J . A . M. <jam...@ab...> - 2001-10-21 23:47:37
|
Hi. I am now in the process of building a beowulf system with dual SMP boxes. System is made with a front end and five dual boxes. And the question is...I see two policies to follow: - Run one bpslave on each node, so system looks lie a 5-cpu one, and you could run two jobs in parallel with no overhead - Run two slaves on each box, to have the system looke like a 10-cpu box. How can the second setup be configured ? TIA -- J.A. Magallon # Let the source be with you... mailto:jam...@ab... Mandrake Linux release 8.2 (Cooker) for i586 Linux werewolf 2.4.12-ac3-beo #4 SMP Thu Oct 18 21:52:06 CEST 2001 i686 |
|
From: Peter B. <pz...@sc...> - 2001-10-20 02:49:18
|
On Thu, 18 Oct 2001, Erik Arjan Hendriks wrote: > This is likely some library changes wrt to nodes which are not > "up". Basically, the library used to check node state and if the > state was not "up" it would use different rfork() flags which > transported all libraries regardless of whether or not they were on > the library list. > I assume that the code that you are refering to is the function default_flags in clients/bproc.c > I should have made that an #ifdef or something I guess... at least > until I get LANL beoboot out the door. That should hopefully be > happening in the next few days since I'm pretty sure I got all the > paper required for that done. *crosses fingers* > I think the attached patch should fix this, but I haven't tested it yet. Thanks. Peter |
|
From: Erik A. H. <er...@he...> - 2001-10-20 00:05:16
|
On Fri, Oct 19, 2001 at 01:55:36AM -0400, Rob Latham wrote: > On Thu, Oct 18, 2001 at 08:54:44PM -0600, Erik Arjan Hendriks wrote: > > > I should have made that an #ifdef or something I guess... at least > > until I get LANL beoboot out the door. That should hopefully be > > happening in the next few days since I'm pretty sure I got all the > > paper required for that done. *crosses fingers* > > paper? are you presenting at SC2001 ? No paper... This stuff will be in the LANL booth at SC2001 and ALS though. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: Rob L. <rl...@pl...> - 2001-10-19 05:55:40
|
On Thu, Oct 18, 2001 at 08:54:44PM -0600, Erik Arjan Hendriks wrote: > I should have made that an #ifdef or something I guess... at least > until I get LANL beoboot out the door. That should hopefully be > happening in the next few days since I'm pretty sure I got all the > paper required for that done. *crosses fingers* paper? are you presenting at SC2001 ? ==rob -- [ Rob Latham <rl...@pl...> Developer, Admin, Alchemist ] [ Paralogic Inc. - www.plogic.com ] [ ] [ EAE8 DE90 85BB 526F 3181 1FCF 51C4 B6CB 08CC 0897 ] |
|
From: Erik A. H. <er...@he...> - 2001-10-19 05:17:50
|
On Fri, Oct 19, 2001 at 01:08:01AM -0400, henken wrote: > Is the pid masquerading visible somewhere in /proc in 3.x on the master? > Or is it just on the slave? PID Masquerading only happens on the slave. Since BProc 3.x those PID translations are applied to what an application sees in /proc (on the slave node) as well. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: henken <he...@se...> - 2001-10-19 05:08:07
|
Oh--it was my fault, I deleted a line when hand patching the rejects. On Thu, 18 Oct 2001, Erik Arjan Hendriks wrote: > On Thu, Oct 18, 2001 at 07:43:18PM -0400, henken wrote: > > > > apparently the bproc modules would like __all__ of the symbols it needs > > exported. New patch is here: > > > > http://reno.cis.upenn.edu/~henken/patches > > What symbols was it complaining about? I suspect this is a Red Hat > kernel problem since I don't see any problems. (at least thru 2.4.10) > > > P.S Any plans to make the bproc stuff move to using the GPL symbols? > > Not at this point. It's not critical and it would needlessly hurt > backward compatibility right now. When I generate a patch that will > only work on linux 2.4.12 or later, I'll think about it. The 2.4.10 > patch appears to apply cleanly but I haven't tested it. > > I really don't think that this kind of thing is a big deal. Having > the system automagically bitch or not doesn't change the license on > BProc. There's virtually no BProc code *in* the kernel anyway. If > somebody wrote non-GPL modules to use those hooks, it would amount to > reimplementing the whole thing. > Gotcha -- wasnt sure if that was the case or not. > - Erik > One more question: Is the pid masquerading visible somewhere in /proc in 3.x on the master? Or is it just on the slave? Nic -- Nicholas Henke Undergraduate - SEAS '02 Liniac Project - University of Pennsylvania http://clubmask.sourceforge.net ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ work: 215-873-5149 cell/home: 215-681-2705 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ There's nothing like good food, good beer, and a bad girl. |
|
From: Erik A. H. <er...@he...> - 2001-10-19 03:14:19
|
On Thu, Oct 18, 2001 at 07:43:18PM -0400, henken wrote: > > apparently the bproc modules would like __all__ of the symbols it needs > exported. New patch is here: > > http://reno.cis.upenn.edu/~henken/patches What symbols was it complaining about? I suspect this is a Red Hat kernel problem since I don't see any problems. (at least thru 2.4.10) > P.S Any plans to make the bproc stuff move to using the GPL symbols? Not at this point. It's not critical and it would needlessly hurt backward compatibility right now. When I generate a patch that will only work on linux 2.4.12 or later, I'll think about it. The 2.4.10 patch appears to apply cleanly but I haven't tested it. I really don't think that this kind of thing is a big deal. Having the system automagically bitch or not doesn't change the license on BProc. There's virtually no BProc code *in* the kernel anyway. If somebody wrote non-GPL modules to use those hooks, it would amount to reimplementing the whole thing. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: Erik A. H. <er...@he...> - 2001-10-19 02:54:54
|
On Thu, Oct 18, 2001 at 02:02:33PM -0400, Peter Bowen wrote: > I'm using BProc 3.1.1 with Red Hat Linux 7.2, and am getting -EIO from > bproc_rfork when I try to use bpcp. I patched the glibc with both > patches, so I'm wondering whether this is the cause, or if it is a known > bug. bpsh works just fine, and the nodes are in the error state > (correct, because the node_up script is failing on bpcp). I can > manually make the node up, and is seems to work. Is this accompanied by messages on the slave console that look something like "vmadump: mmap failed: <library name>"? Presuming this is the case.... This is likely some library changes wrt to nodes which are not "up". Basically, the library used to check node state and if the state was not "up" it would use different rfork() flags which transported all libraries regardless of whether or not they were on the library list. With the beoboot that's out there now this was necessary because libraries aren't present at boot time. You get and EIO if the undump fails due missing libraries or whatever. Since new boot support/library caching stuff can fetch libraries on demand, the BProc library doesn't do this check anymore. If bpsh still works, you must still be using the first dynamic linker hack which still does that check. I should have made that an #ifdef or something I guess... at least until I get LANL beoboot out the door. That should hopefully be happening in the next few days since I'm pretty sure I got all the paper required for that done. *crosses fingers* - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: Erik A. H. <er...@he...> - 2001-10-19 02:41:44
|
On Thu, Oct 18, 2001 at 10:41:23AM -0700, Shane Walton wrote: > Can bproc do any magic as to mask the IP of a process > running on a given node to the IP of the front end > node? I guess I am asking, when a process is excuted > on a remote node, which was originally executed on the > front-end, will it inherit the IP of the front-end. > If it doesn't, does anyone know a way to make a > cluster look like it has only one IP? There's no such magic in BProc and deliberately so. That particular flavor of magic is pretty difficult to do and would almost certainly be expensive. I've avoided doing anything that would cause any kind of performance hit on your average MPI application. That basically means no added over head on any file or socket IO which means handling it all locally on the node. If you're worried about what things look like to the outside world, you can make it look like outgoing connections are coming from the same place by using the front end as a router and setting up NAT. Linux virtual server might be useful for the reverse. - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: henken <he...@se...> - 2001-10-18 23:43:20
|
apparently the bproc modules would like __all__ of the symbols it needs exported. New patch is here: http://reno.cis.upenn.edu/~henken/patches Nic P.S Any plans to make the bproc stuff move to using the GPL symbols? -- Nicholas Henke Undergraduate - SEAS '02 Liniac Project - University of Pennsylvania http://clubmask.sourceforge.net ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ work: 215-873-5149 cell/home: 215-681-2705 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ There's nothing like good food, good beer, and a bad girl. |
|
From: henken <he...@se...> - 2001-10-18 23:00:46
|
the spec file in the tarball needs to have the dirs /usr/share/man/man1,man2,man8 created. The spec file failes if not Nic -- Nicholas Henke Undergraduate - SEAS '02 Liniac Project - University of Pennsylvania http://clubmask.sourceforge.net ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ work: 215-873-5149 cell/home: 215-681-2705 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ There's nothing like good food, good beer, and a bad girl. |
|
From: henken <he...@se...> - 2001-10-18 18:33:30
|
Here is the bproc patch for 2.4.9 modified for the redhat 2.4.9 version of that kernel. RPMS/SRPMS for redhat 7.1 with all of the updates are also complied. ( They will be available soon ). These rpms also require an update to e2fsprogs-1.23, filesystem-2.1.6, setup-2.5.7, modutils-2.4.10, and mkinitrd-3.2.6. All of these can be found here: http://reno.cis.upenn.edu/~henken/RPMS http://reno.cis.upenn.edu/~henken/SRPMS If I have screwed something up please let me know Nic -- Nicholas Henke Undergraduate - SEAS '02 Liniac Project - University of Pennsylvania http://clubmask.sourceforge.net ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ work: 215-873-5149 cell/home: 215-681-2705 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ There's nothing like good food, good beer, and a bad girl. |
|
From: Peter B. <pz...@da...> - 2001-10-18 18:02:42
|
I'm using BProc 3.1.1 with Red Hat Linux 7.2, and am getting -EIO from bproc_rfork when I try to use bpcp. I patched the glibc with both patches, so I'm wondering whether this is the cause, or if it is a known bug. bpsh works just fine, and the nodes are in the error state (correct, because the node_up script is failing on bpcp). I can manually make the node up, and is seems to work. I haven't run the tests yet to verify that everything else is working, but I was hoping that someone else might have seen this. Thanks. Peter |
|
From: Erik A. H. <er...@he...> - 2001-10-14 03:44:11
|
On Tue, Oct 09, 2001 at 11:21:21AM +0200, Janez Pers wrote: > Another question: it seems everything is working now, but only if > I login as root (not surprisingly, as the whole installation was done > with root permisions). > > Somewhere in the docs it was stated that the whole system should work > even without root permisions. However, executing /usr/sbin/bpmaster > as ordinary user gives following error message: The system works *for* non-root users but it cannot (well, should not...) be run by non-root users. What this means is root needs to run the daemons but after that anybody can call bproc_move(), etc. > mmap: permission denied. > > This is not critical, as I intend to load the modules and the master > daemon during the system startup. > > However, running /usr/sbin/bpslave <ip> <port> refused to run with: > > bindresvport: permission denied (ignoring) > /dev/bproc: permission denied > > After I changed the permissions of /dev/bproc with chmod g+rw /dev/bproc > and chgrp /dev/bproc users (I don't know, was that a very stupid thing?) > > the bpslave still refuses to run with: > > bindresvport: permission denied (ignoring) > /dev/bproc: operation not permitted All these problems are related to not being root. Trying to chmod things to the point where anybody can run them is a very bad idea. It creates a huge gaping security hole. (newer revs of bproc remove /dev/bproc and doing that kind of thing isn't even an option...) - Erik -- Erik Arjan Hendriks Printed On 100 Percent Recycled Electrons er...@he... Contents may settle during shipment |
|
From: Anant <an...@ya...> - 2001-10-11 05:31:12
|
Hello,
Being a newbie myself I faced lots of problems while setting up
BProc for my Degree Project. I have written down steps to do the
installation for bproc and make it up and running. I hope this is of
help. Please do correct me if I am incorrect at any place.
please follow the following link to access the document
http://anant_gupta.tripod.com/projects/IT/bproc_installation.htm
I sent a previous mail along with the document attached but it is
awaiting moderators approval. Kindly ignore it in case that mail is
also sent.
Thanks.
Anant
> Message: 4
> Date: Mon, 8 Oct 2001 09:43:53 -0600
> From: Erik Arjan Hendriks <er...@he...>
> To: Janez Pers <jan...@fe...>
> Cc: bpr...@li...
> Subject: Re: [BProc] beowulf demo cluster
> > The bproc documentation on sourceforge is ok, although many steps
> > in installation are missing, so can anyone write down step-by-step
> > instructions which make options should be executed and in which
> > order (sort of "bproc install for dummies").
>
> It's somewhere on my very long todo list. If somebody by any chance
> already done this, I'd love to post it up there.
>
> - Erik
=====
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Anant Kumar Gupta
III A - 3,Nehru Nagar,
Ghaziabad. U.P.201 001. Phone : 4720720, 4714714
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Perfection is my goal and
only Excellence will be Tolerated.
__________________________________________________
Do You Yahoo!?
Make a great connection at Yahoo! Personals.
http://personals.yahoo.com
|
|
From: Janez P. <jan...@fe...> - 2001-10-09 09:21:30
|
Erik Arjan Hendriks wrote: > > > > Also, insmod /usr/src/bproc/kernel/bproc.o says: > > Cannot find module bproc.o ... BUT IT IS there! > > I have no answer for this one. Try using strace on insmod and see > where it's getting confused. I started solving the problems from the end to the beginning, and everything worked... i did not execute "make mods", therefore bproc.o was not quite what it should be... after make mods the module of course mounts ok and the device /dev/bproc gets alive... Another question: it seems everything is working now, but only if I login as root (not surprisingly, as the whole installation was done with root permisions). Somewhere in the docs it was stated that the whole system should work even without root permisions. However, executing /usr/sbin/bpmaster as ordinary user gives following error message: mmap: permission denied. This is not critical, as I intend to load the modules and the master daemon during the system startup. However, running /usr/sbin/bpslave <ip> <port> refused to run with: bindresvport: permission denied (ignoring) /dev/bproc: permission denied After I changed the permissions of /dev/bproc with chmod g+rw /dev/bproc and chgrp /dev/bproc users (I don't know, was that a very stupid thing?) the bpslave still refuses to run with: bindresvport: permission denied (ignoring) /dev/bproc: operation not permitted Now I am a bit confused... Janez. |
|
From: Erik A. H. <er...@he...> - 2001-10-08 15:43:57
|
On Mon, Oct 08, 2001 at 05:37:42PM +0200, Janez Pers wrote: > However, bpmaster refuses to run, it complains that > > /dev/bproc: no such device. You'll get that message until you actually install the bproc kernel module. bproc.o is the device in question and if the module isn't loaded, the device isn't there. > I find it utterly stupid, as the entry /dev/bproc is there, it > was created by the install script... > > Also, insmod /usr/src/bproc/kernel/bproc.o says: > Cannot find module bproc.o ... BUT IT IS there! I have no answer for this one. Try using strace on insmod and see where it's getting confused. > The bproc documentation on sourceforge is ok, although many steps > in installation are missing, so can anyone write down step-by-step > instructions which make options should be executed and in which > order (sort of "bproc install for dummies"). It's somewhere on my very long todo list. If somebody by any chance already done this, I'd love to post it up there. - Erik |
|
From: Janez P. <jan...@fe...> - 2001-10-08 15:37:42
|
Hi all. I am trying to set up simple beowulf (1 master, 1 node) system, for demonstrational (educational) purposes. Therefore, I do not need real supercomputing power :> just the ability to show to the students the parallel processing with bproc_rfork, as part of exercise for the course "Parallel systems" which is teached by my supervisor. In ideal case I would set up the student boxes in our lab to work as a nodes, so the students do some experimenting... I've been trying every possible combinations of bproc distributions, kernels and various linux distributions (yes, and I even tried scyld beowulf - did not work out) for one week now an I am really frustrated. (yes, Vmware rocks and is the only bright point there, otherwise I would quit it long before) The point where I really stopped and I see no way out is combination of kernel 2.2.18 and bproc 2.2.2 on redhat 7.0 * it patches ok * kernel compiles ok * after installation kernel works, but complains it cannot find module dependencies during boot. However, network card works without problems, and is compiled as a module) * bproc compiles ok (make) * it installs (make install) (I see no obvious warnings) However, bpmaster refuses to run, it complains that /dev/bproc: no such device. I find it utterly stupid, as the entry /dev/bproc is there, it was created by the install script... Also, insmod /usr/src/bproc/kernel/bproc.o says: Cannot find module bproc.o ... BUT IT IS there! What I am doing wrong ? I know I will be getting responses that i am using old version of bproc, however, I suspect that is not the problem, I think I screwed up somewhere along setup and new versions of bproc/kernel would make no difference. The bproc documentation on sourceforge is ok, although many steps in installation are missing, so can anyone write down step-by-step instructions which make options should be executed and in which order (sort of "bproc install for dummies"). Thanks in advance, Janez. |
1488 messages has been excluded from this view by a project administrator.
