bproc-users — General discussion about BProc.

[BProc] unified parallel C

[Daniel G. <dg...@cp...>]

Hi All,

Is anybody using Unified Parallel C on BProc?  Has anybody
implemented GASNet on BProc?

Regards,
Daniel
-- 

Dr. Daniel Gruner                        dg...@ch...
Dept. of Chemistry                       dan...@ut...
University of Toronto                    phone:  (416)-978-8689
80 St. George Street                     fax:    (416)-978-5325
Toronto, ON  M5S 3H6, Canada             finger for PGP public key

Re: [BProc] performance issues with Myrinet : latency

[Maurice H. <ma...@ha...>]

Florent Calvayrac wrote:

>This is probably a well know issue but since we have installed
>Clustermatic  (one version away) the latency of our Myrinet network
>has gone way up, from 3-8 microseconds to about 100
>for any type of MPI call. I am sure not to have
>installed the TCP/IP version (since only two
>jobs per card seem to be running at the same time at maximum)
>and I suspect the problem comes from dynamic routing.
>
>I tried to download and compile the router given on the clustermatic
>home page but fail because of a missing "gm_undocumented_calls.h"
>and other errors.
>  
>

I suggest you should try running "MX"" drivers rather than the older gmpich

--

With our best regards,

Maurice W. Hilarius
Hard Data Ltd

Re: [BProc] clustermatic: node dies with signal 11

[Ron S. <rse...@ha...>]

>> nodeup    :   No premove function for nodeinfo
               ^^^^^^^^^^^^^^^^^^^^^^^

>I believe I've seen this before when the stage2 image and front-end
>had bproc mismatches, i.e. I had {re-}built bproc on the
> front-end w/o generating a new stage2 image.  Could that
> be possible in your case?

I built everything 'stock' from the clustermatic rpms and even
double-checked by rebuilding the stage 2 after making any config
changes, and putting the new stage 2 onto the pxe server. I'm building
the stage 2 with the command:
[root@novac root]# beoboot -2 -i -v -o /tmp/phase2

On 11/17/05, Joshua Aune <lu...@ln...> wrote:
> On Wed, 2005-11-16 at 23:05 -0600, Ron Senykoff wrote:
>
> Do you have anything coming out of the serial console (or vga console if
> not configured with serial) on the node?

Yes... any node I try (I've tried several to make sure it wasn't bad
hardware) comes up : (.219 is the server, .220 starts the node range)
boot: Server IP address: 192.168.46.219
boot: My IP address    : 192.168.26.222
boot: starting bpslave: bpslave -d -i -v 192.168.46.219 2223
bpslave: connecting to 192.168.46.219:2223
bpslave: IO daemon started; pid=3D17
bpslave: connection to 192.168.46.219:2223 up and running
bpslave: Setting node number to 2

I have now moved my install back to clustermatic 4 on Redhat 9 (hoping
that i386 builds may remedy the situation) and I get the exact same
message on the node, but I get 'Signal 4' instead of 'Signal 11' in
/var/log/clustermatic/node.X:

[root@novac root]# tail /var/log/clustermatic/node.2
vmadlib   :     loaded /lib/libnss_bproc.so.2 (size=3D25043;id=3D0,0;mode=
=3D100755)
vmadlib   :     loaded /usr/lib/libbproc.so.4.0.0
(size=3D21388;id=3D0,0;mode=3D100755)
vmadlib   :     loaded /usr/lib/libstdc++.so.5.0.3
(size=3D710608;id=3D0,0;mode=3D100755)
nodeup    :   Plugin vmadlib returned status 0 (ok)
nodeup    :   No premove function for nodeinfo
nodeup    : Starting 1 child processes.
nodeup    : Finished creating child processes.
nodeup    : I/O error talking to child
nodeup    : Child process for node 2 died with signal 4
nodeup    : Node setup returned status 1

This was another complete from scratch rebuild. Put Redhat 9 minimal +
dev packages on laptop. Installed clustermatic. Reboot into new
kernel. Build phase 2, put that on PXE server...

Any help is much appreciated. TIA!
-Ron

Re: [BProc] clustermatic: node dies with signal 11

[Joshua A. <lu...@ln...>]

On Wed, 2005-11-16 at 23:05 -0600, Ron Senykoff wrote:

Do you have anything coming out of the serial console (or vga console if
not configured with serial) on the node?

> Could it be the difference between the laptop and the nodes? They are
> both i586 correct? Any help is greatly appreciated.
> 
> [root@novac root]# tail /var/log/clustermatic/node.0
> vmadlib   :     loaded /lib/libnss_bproc.so.2 (size=44922;id=0,0;mode=100755)
> vmadlib   :     loaded /usr/lib/libbproc.so.4.0.0
> (size=40697;id=0,0;mode=100755)
> vmadlib   :     loaded /usr/lib/libstdc++.so.5.0.5
> (size=732372;id=0,0;mode=100755)
> nodeup    :   Plugin vmadlib returned status 0 (ok)
> nodeup    :   No premove function for nodeinfo
> nodeup    : Starting 1 child processes.
> nodeup    : Finished creating child processes.
> nodeup    : I/O error talking to child
> nodeup    : Child process for node 0 died with signal 11
> nodeup    : Node setup returned status 1
> 
> [root@novac root]# cat /etc/clustermatic/config
> interface eth0
> 
> #master 192.168.46.219
> master novac
> 
> iprange 0 192.168.46.20 192.168.46.35   # Nodes 0-8 have addresses
> from this range.
> 
> bootfile /var/clustermatic/boot.img
> 
> librariesfrombinary /bin/sleep /bin/ps /bin/ping /bin/ls # get libc,resolver
> libraries /usr/lib/libstdc++*   /usr/lib64/libstdc++*    # C++ support
> libraries /usr/lib/libbproc.so* /usr/lib64/libbproc.so*  # BProc, of course.
> libraries /lib/libnss_bproc*    /lib64/libnss_bproc*     # BProc resolver
> 
> node 0 00:60:EF:21:61:AF
> node  00:60:EF:21:63:16
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by the JBoss Inc.  Get Certified Today
> Register for a JBoss Training Course.  Free Certification Exam
> for All Training Attendees Through End of 2005. For more info visit:
> http://ads.osdn.com/?ad_idv28&alloc_id845&op=click
> _______________________________________________
> BProc-users mailing list
> BPr...@li...
> https://lists.sourceforge.net/lists/listinfo/bproc-users

[BProc] performance issues with Myrinet : latency

[Florent C. <Flo...@un...>]

Hi to all

This is probably a well know issue but since we have installed
Clustermatic  (one version away) the latency of our Myrinet network
has gone way up, from 3-8 microseconds to about 100
for any type of MPI call. I am sure not to have
installed the TCP/IP version (since only two
jobs per card seem to be running at the same time at maximum)
and I suspect the problem comes from dynamic routing.

I tried to download and compile the router given on the clustermatic
home page but fail because of a missing "gm_undocumented_calls.h"
and other errors.

 Is anyone aware of this problem ?
Is it worth upgrading to the last version of clustermatic
as this would probably involve several days of work ?
(we have some local specificities and all our programs run fine excepted 
for one which is
latency bound)



thanks in advance


-- 
Florent Calvayrac  | Tel : 06 64 31 43 86  | http://www.jackywulf.com
Directeur du SC Informatique Ressources Num.  de l'Universite du Maine
Lab. de Physique de l'Etat Condense UMR-CNRS 6087 
Inst. de Rech. en Ingenierie Molec. et Matx Fonctionnels  FR CNRS 2575 

Re: [BProc] clustermatic: node dies with signal 11

[Daryl W. G. <dw...@la...>]

> Date: Wed, 16 Nov 2005 23:05:16 -0600
> From: Ron Senykoff <rse...@gm...>
> To: bpr...@li...
> Subject: [BProc] clustermatic: node dies with signal 11
> 
> I am working on building a cluster of thin-clients for demo purposes
> (32 nodes). I'm truly stuck, and these 8-week courses are killing us
> grad students to get solid projects completed...
> 
> Using PXE the nodes load stage2 of clustermatic fine. However, the
> nodes can't connect correctly to the server (a laptop).
> 
> A bit of info on the hardware:
> laptop = PII w/ 64MB of RAM (I reduced it to be identical to the clients)
> nodes = AMD K6-2 w/ 64 MB of RAM + DiskOnChip (we want to put the
> successful build onto that so no network booting)
> Could it be the difference between the laptop and the nodes? They are
> both i586 correct? Any help is greatly appreciated.
> 
> [root@novac root]# tail /var/log/clustermatic/node.0
> vmadlib   :     loaded /lib/libnss_bproc.so.2 (size=44922;id=0,0;mode=100755)
> vmadlib   :     loaded /usr/lib/libbproc.so.4.0.0
> (size=40697;id=0,0;mode=100755)
> vmadlib   :     loaded /usr/lib/libstdc++.so.5.0.5
> (size=732372;id=0,0;mode=100755)
> nodeup    :   Plugin vmadlib returned status 0 (ok)
> nodeup    :   No premove function for nodeinfo
                ^^^^^^^^^^^^^^^^^^^^^^^

I believe I've seen this before when the stage2 image and front-end had bproc
mismatches, i.e. I had {re-}built bproc on the front-end w/o generating a
new stage2 image.  Could that be possible in your case?

Daryl

> nodeup    : Starting 1 child processes.
> nodeup    : Finished creating child processes.
> nodeup    : I/O error talking to child
> nodeup    : Child process for node 0 died with signal 11
> nodeup    : Node setup returned status 1
> 
> [root@novac root]# cat /etc/clustermatic/config
> interface eth0
> 
> #master 192.168.46.219
> master novac
> 
> iprange 0 192.168.46.20 192.168.46.35   # Nodes 0-8 have addresses
> from this range.
> 
> bootfile /var/clustermatic/boot.img
> 
> librariesfrombinary /bin/sleep /bin/ps /bin/ping /bin/ls # get libc,resolver
> libraries /usr/lib/libstdc++*   /usr/lib64/libstdc++*    # C++ support
> libraries /usr/lib/libbproc.so* /usr/lib64/libbproc.so*  # BProc, of course.
> libraries /lib/libnss_bproc*    /lib64/libnss_bproc*     # BProc resolver
> 
> node 0 00:60:EF:21:61:AF
> node  00:60:EF:21:63:16

[BProc] clustermatic: node dies with signal 11

[Ron S. <rse...@gm...>]

I am working on building a cluster of thin-clients for demo purposes
(32 nodes). I'm truly stuck, and these 8-week courses are killing us
grad students to get solid projects completed...

Using PXE the nodes load stage2 of clustermatic fine. However, the
nodes can't connect correctly to the server (a laptop).

A bit of info on the hardware:
laptop =3D PII w/ 64MB of RAM (I reduced it to be identical to the clients)
nodes =3D AMD K6-2 w/ 64 MB of RAM + DiskOnChip (we want to put the
successful build onto that so no network booting)
Could it be the difference between the laptop and the nodes? They are
both i586 correct? Any help is greatly appreciated.

[root@novac root]# tail /var/log/clustermatic/node.0
vmadlib   :     loaded /lib/libnss_bproc.so.2 (size=3D44922;id=3D0,0;mode=
=3D100755)
vmadlib   :     loaded /usr/lib/libbproc.so.4.0.0
(size=3D40697;id=3D0,0;mode=3D100755)
vmadlib   :     loaded /usr/lib/libstdc++.so.5.0.5
(size=3D732372;id=3D0,0;mode=3D100755)
nodeup    :   Plugin vmadlib returned status 0 (ok)
nodeup    :   No premove function for nodeinfo
nodeup    : Starting 1 child processes.
nodeup    : Finished creating child processes.
nodeup    : I/O error talking to child
nodeup    : Child process for node 0 died with signal 11
nodeup    : Node setup returned status 1

[root@novac root]# cat /etc/clustermatic/config
interface eth0

#master 192.168.46.219
master novac

iprange 0 192.168.46.20 192.168.46.35   # Nodes 0-8 have addresses
from this range.

bootfile /var/clustermatic/boot.img

librariesfrombinary /bin/sleep /bin/ps /bin/ping /bin/ls # get libc,resolve=
r
libraries /usr/lib/libstdc++*   /usr/lib64/libstdc++*    # C++ support
libraries /usr/lib/libbproc.so* /usr/lib64/libbproc.so*  # BProc, of course=
.
libraries /lib/libnss_bproc*    /lib64/libnss_bproc*     # BProc resolver

node 0 00:60:EF:21:61:AF
node  00:60:EF:21:63:16

Re: [BProc] Clustermatic 5 Queueing system

[Joshua A. <lu...@ln...>]

I know that LSF has been integrated to work with BProc as well, but I
have just used bjs.

On Mon, 2005-11-14 at 11:51 -0500, Daniel Gruner wrote:
> Hi Patrice,
> 
> I use bjs, which is the batch scheduler that comes with cm5.  I use
> version 1.6.  It is pretty simple minded, but it works quite well.
> The only issue with it - and it may not be an issue for you at all -
> is that the batch system gives you nodes, rather than processors.
> In other words, if your nodes are dual-cpu you need to run 2 processes
> on the node in order to take advantage of the machine.  We do this for
> serial jobs by submitting 2 processes in the script that the user
> submits to bjs.  For mpi codes this is not a problem at all - you
> simply tell mpirun to pack the processes on the nodes (you need to use
> the mpirun that is supplied with clustermatic, since it is the only
> one that is bproc-aware).
> 
> It would be really nice if someone went through the trouble of porting
> SGE or OpenPBS or something else to bproc, but in the meantime bjs 
> does the job.
> 
> Regards,
> Daniel
> 
> On Mon, Nov 14, 2005 at 11:18:59AM +0100, patrice descourt wrote:
> > Hello
> > 
> > I use Clustermatic 5 right now on a 16 diskless nodes cluster and look 
> > for a good Queueing and Batch  job submission system to run with 
> > Clustermatic 5
> > 
> > Does anyone know one ?
> > 
> > Thx a lot
> > 
> >  P.Descourt
> > 
> > U650 INSERM,
> > Laboratoire de Traitement d'Information Medicale (LaTIM),
> > Equipe 'Quantification en Tomographie d'Emission',
> > CHU Morvan,
> > 5 avenue Foch,
> > 29609 Brest, France
> > 
> > 
> > 
> > 
> > -------------------------------------------------------
> > SF.Net email is sponsored by:
> > Tame your development challenges with Apache's Geronimo App Server. Download
> > it for free - -and be entered to win a 42" plasma tv or your very own
> > Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> > _______________________________________________
> > BProc-users mailing list
> > BPr...@li...
> > https://lists.sourceforge.net/lists/listinfo/bproc-users
> 

Re: [BProc] Clustermatic 5 Queueing system

[Daniel G. <dg...@cp...>]

Hi Patrice,

I use bjs, which is the batch scheduler that comes with cm5.  I use
version 1.6.  It is pretty simple minded, but it works quite well.
The only issue with it - and it may not be an issue for you at all -
is that the batch system gives you nodes, rather than processors.
In other words, if your nodes are dual-cpu you need to run 2 processes
on the node in order to take advantage of the machine.  We do this for
serial jobs by submitting 2 processes in the script that the user
submits to bjs.  For mpi codes this is not a problem at all - you
simply tell mpirun to pack the processes on the nodes (you need to use
the mpirun that is supplied with clustermatic, since it is the only
one that is bproc-aware).

It would be really nice if someone went through the trouble of porting
SGE or OpenPBS or something else to bproc, but in the meantime bjs 
does the job.

Regards,
Daniel

On Mon, Nov 14, 2005 at 11:18:59AM +0100, patrice descourt wrote:
> Hello
> 
> I use Clustermatic 5 right now on a 16 diskless nodes cluster and look 
> for a good Queueing and Batch  job submission system to run with 
> Clustermatic 5
> 
> Does anyone know one ?
> 
> Thx a lot
> 
>  P.Descourt
> 
> U650 INSERM,
> Laboratoire de Traitement d'Information Medicale (LaTIM),
> Equipe 'Quantification en Tomographie d'Emission',
> CHU Morvan,
> 5 avenue Foch,
> 29609 Brest, France
> 
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App Server. Download
> it for free - -and be entered to win a 42" plasma tv or your very own
> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> _______________________________________________
> BProc-users mailing list
> BPr...@li...
> https://lists.sourceforge.net/lists/listinfo/bproc-users

-- 

Dr. Daniel Gruner                        dg...@ch...
Dept. of Chemistry                       dan...@ut...
University of Toronto                    phone:  (416)-978-8689
80 St. George Street                     fax:    (416)-978-5325
Toronto, ON  M5S 3H6, Canada             finger for PGP public key

[BProc] Clustermatic 5 Queueing system

[patrice d. <pat...@un...>]

Hello

I use Clustermatic 5 right now on a 16 diskless nodes cluster and look 
for a good Queueing and Batch  job submission system to run with 
Clustermatic 5

Does anyone know one ?

Thx a lot

 P.Descourt

U650 INSERM,
Laboratoire de Traitement d'Information Medicale (LaTIM),
Equipe 'Quantification en Tomographie d'Emission',
CHU Morvan,
5 avenue Foch,
29609 Brest, France

Re: [BProc] Re: Clustermatic"s kernel-2.6.9-cm46: NMI lockup on CPU1, CPU3

[Joshua A. <lu...@ln...>]

On Fri, 2005-11-11 at 16:24 -0500, Jeff Palmucci wrote:
> I'm having the exact same problem as mentioned below (previously on this 
> list). Was there any resolution?

I have some patches for BProc4-pre and linux-2.6.14...  Maybe that will
help.  I am hoping to get these put on the sourceforge page eventually.
Anyone that is interested contact me off list for a pre-release.

> -----------------------------------------
> 
> From: Motu <motudai@gm...>
> Re: Clustermatic"s kernel-2.6.9-cm46: NMI lockup on CPU1, CPU3 
> 2005-08-09 16:56
> 
> 
> 
> Hello,
> I am working with a machine with 2 250 gb SATA harddrives, 2 dual-core
> opteron 2.2 processors, 2 broadcom netxtreme 5704 gigabit ethernet network
> cards, and a tyan s2881 motherboard.
> When I try to boot with the kernel version 2.6.9-cm46 (the one provided 
> with
> clustermatic 5)--as opposed to the 2.6.11-FC4smp that came with the Fedora
> Core 4 system, I get NMI lockup messages on both the CPUs.
> The following is what the tail-end of hte message looks like:
> 
>  > Code: 80 3f 00 7e f9 e9 8b fd ff ff e8 1e bf ec ff e9 a1 fd ff ff
> 
>  Console shuts up...
>  >
>  NMI Watchdog detected LOCKUP on CPU1, registers:
>  > CPU1
>  > Modules Linked in:
>  > Pid:0, comm:swapper No tainted 2.6.9-cm46
>  > RIP:0010:[] NMI Watchdog detected LOCKUP on CPU3,
>  > registers:
>  > CPU3
>  > Modules Linked in:
>  > Pid:0, comm:swapper No tainted 2.6.9-cm46
>  > RIP:0010:[]{schedulertick+385}
> 
> 
> Does anyone know the cause of this problem? Anyway to deduce it?
> The web seems to be useless: everyone with a NMI lockup fixed the problem
> with a newer kernel, but in my case, I need to use clustermatic"s 
> kernel. Is
> there any way out?
> 
> Prabhas
> 
>            
> 
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App Server. Download
> it for free - -and be entered to win a 42" plasma tv or your very own
> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> _______________________________________________
> BProc-users mailing list
> BPr...@li...
> https://lists.sourceforge.net/lists/listinfo/bproc-users

[BProc] Re: Clustermatic"s kernel-2.6.9-cm46: NMI lockup on CPU1, CPU3

[Jeff P. <jpa...@ma...>]

I'm having the exact same problem as mentioned below (previously on this 
list). Was there any resolution?

-----------------------------------------

From: Motu <motudai@gm...>
Re: Clustermatic"s kernel-2.6.9-cm46: NMI lockup on CPU1, CPU3 
2005-08-09 16:56



Hello,
I am working with a machine with 2 250 gb SATA harddrives, 2 dual-core
opteron 2.2 processors, 2 broadcom netxtreme 5704 gigabit ethernet network
cards, and a tyan s2881 motherboard.
When I try to boot with the kernel version 2.6.9-cm46 (the one provided 
with
clustermatic 5)--as opposed to the 2.6.11-FC4smp that came with the Fedora
Core 4 system, I get NMI lockup messages on both the CPUs.
The following is what the tail-end of hte message looks like:

 > Code: 80 3f 00 7e f9 e9 8b fd ff ff e8 1e bf ec ff e9 a1 fd ff ff

 Console shuts up...
 >
 NMI Watchdog detected LOCKUP on CPU1, registers:
 > CPU1
 > Modules Linked in:
 > Pid:0, comm:swapper No tainted 2.6.9-cm46
 > RIP:0010:[] NMI Watchdog detected LOCKUP on CPU3,
 > registers:
 > CPU3
 > Modules Linked in:
 > Pid:0, comm:swapper No tainted 2.6.9-cm46
 > RIP:0010:[]{schedulertick+385}


Does anyone know the cause of this problem? Anyway to deduce it?
The web seems to be useless: everyone with a NMI lockup fixed the problem
with a newer kernel, but in my case, I need to use clustermatic"s 
kernel. Is
there any way out?

Prabhas

           

[BProc] compiling clubmask for BPROC

[patrice d. <pat...@un...>]

Hello

I setup an AMD64 cluster with 27 AMD64 DISKLESS nodes with Clustermatic=20
5 and intend to use Clubmask 06b2 as a scheduler with Maui and Ganglia

but when I try to compile the distribution via

python setup.py install

I get a lot of error messages in the compiling phase :

gcc -DNDEBUG -g -O3 -Wall -Wstrict-prototypes -fPIC=20
-I/usr/local/include/python2.2 -c src/bproc/bprocmodule.c -o=20
build/temp.linux-i686-2.2/bprocmodule.o -ggdb
src/bproc/bprocmodule.c: In function `bp_masteraddr':
src/bproc/bprocmodule.c:330: attention : implicit declaration of=20
function `bproc_masteraddr'
src/bproc/bprocmodule.c: In function `bp_nodeinfo':
src/bproc/bprocmodule.c:462: erreur: incompatible type for argument 1 of=20
`makeip'
src/bproc/bprocmodule.c: In function `bp_nodelist':
src/bproc/bprocmodule.c:479: attention : passage de l'argument n=B01 de =AB=
=20
bproc_nodelist =BB d'un type pointeur incompatible
src/bproc/bprocmodule.c:492: erreur: incompatible type for argument 1 of=20
`makeip'
src/bproc/bprocmodule.c: In function `bp_nodenumber':
src/bproc/bprocmodule.c:528: attention : implicit declaration of=20
function `bproc_nodenumber'
src/bproc/bprocmodule.c: In function `bp_nodesetstatus':
src/bproc/bprocmodule.c:601: erreur: `BPROC_NODE_NSTATES' undeclared=20
(first use in this function)
src/bproc/bprocmodule.c:601: erreur: (Each undeclared identifier is=20
reported only once
src/bproc/bprocmodule.c:601: erreur: for each function it appears in.)
src/bproc/bprocmodule.c:606: attention : passage de l'argument n=B02 de =AB=
=20
bproc_nodesetstatus =BB transforme un entier en pointeur sans transtypage
src/bproc/bprocmodule.c: In function `bp_nodestatus':
src/bproc/bprocmodule.c:629: attention : implicit declaration of=20
function `bproc_nodeup'
src/bproc/bprocmodule.c: In function `initbproc':
src/bproc/bprocmodule.c:869: erreur: `bproc_node_down' undeclared (first=20
use in this function)
src/bproc/bprocmodule.c:870: erreur: `bproc_node_unavailable' undeclared=20
(first use in this function)
src/bproc/bprocmodule.c:871: erreur: `bproc_node_error' undeclared=20
(first use in this function)
src/bproc/bprocmodule.c:872: erreur: `bproc_node_up' undeclared (first=20
use in this function)
src/bproc/bprocmodule.c:873: erreur: `bproc_node_reboot' undeclared=20
(first use in this function)
src/bproc/bprocmodule.c:874: erreur: `bproc_node_pwroff' undeclared=20
(first use in this function)
error: command 'gcc' failed with exit status 1


is it because I do not use clustermatic 4 ? or because my python version=20
(2.2.3)  is not the right needed to compile clubmask ??

Thank for your help

P.Descourt

U650 INSERM, Laboratoire de Traitement d'Information Medicale (LATIM)
Equipe 'Quantification en Tomographie d'Emission'
CHU Morvan, B=E2timent 2Bis (I3S), 5 avenue Foch, 29609 Brest, France

Re: [BProc] UP bproc kernels (quick poll)

[Dale H. <ro...@ma...>]

On 2005-11-01 at 16:01, Joshua Aune <lu...@ln...> elucidated:
> Is anyone using the uniprocessor kernels that come with the clustermatic
> distribution?
> 

I wasn't when I was using bproc.  Rolled my own kernels for a SMP box.


--
Dale Harris   
ro...@ma...
/.-)

[BProc] UP bproc kernels (quick poll)

[Joshua A. <lu...@ln...>]

Is anyone using the uniprocessor kernels that come with the clustermatic
distribution?

Just curious.

Josh

Re: [BProc] Mpich on Nimbus

[Joshua A. <lu...@ln...>]

On Fri, 2005-10-21 at 03:20 -0400, Shobana Ravi wrote:
> I am working on the Nimbus 4 system based on Clustermatic on a 15 node 
> cluster of Opterons. I am having intermittent problems running MPI 
> applications on this.
> 
> While trying to run the the ASCI benchmark sPPM, 4 times out of 5, I get 
> errors that look like this :
> 
> p3_16410:  p4_error: interrupt SIGSEGV: 11
>
> 
> If anybody could help me with this problem, or give me pointers as to what 
> to explore/debug, I would greatly appreciate it.


Build your app with debugging symbols, get the task on the node to dump
a core file when it crashes, copy the core file back to the host node
and analyze using gdb.

[BProc] Mpich on Nimbus

[Shobana R. <sh...@cs...>]

I am working on the Nimbus 4 system based on Clustermatic on a 15 node 
cluster of Opterons. I am having intermittent problems running MPI 
applications on this.

While trying to run the the ASCI benchmark sPPM, 4 times out of 5, I get 
errors that look like this :

p3_16410:  p4_error: interrupt SIGSEGV: 11
p3_16410: (23.234375) net_send: could not write to fd=4, errno = 32
p4_16412: (33.445312) net_send: could not write to fd=4, errno = 32
p11_16426: (33.089844) net_send: could not write to fd=4, errno = 32
p2_16408: (33.558594) net_send: could not write to fd=4, errno = 32
p9_16422: (33.207031) net_send: could not write to fd=4, errno = 32
     p4_error: latest msg from perror: Broken pipe
p4_16412:  p4_error: net_send write: -1
     p4_error: latest msg from perror: Broken pipe
p11_16426:  p4_error: net_send write: -1
     p4_error: latest msg from perror: Broken pipe
p2_16408:  p4_error: net_send write: -1
     p4_error: latest msg from perror: Broken pipe
p9_16422:  p4_error: net_send write: -1
p12_16428: (34.140625) net_send: could not write to fd=4, errno = 32
     p4_error: latest msg from perror: Broken pipe
bm_list_16405: (71.308594) net_send: could not write to fd=4, errno = 9
     p4_error: latest msg from perror: Bad file descriptor
bm_list_16405:  p4_error: net_send write: -1

.... and so on. This happens often but not always. Each time the node on 
which the seg fault occurs is different. The exact same code works fine on 
other clusters.

My environment settings are
BEOWULF_JOB_MAP=0:1:2:3:4:5:6:7:8:9:10:11:12:13:14
NP=15

Unfortunately, I dont have the sources for the MPICH. The installed 
version is in a directory named mpich-gnu-lila-p4-1.2.5..10.

I tried compiling MPICH 1.2.7 for this, but I didn't find the patches to 
get it working with Nimbus.

If anybody could help me with this problem, or give me pointers as to what 
to explore/debug, I would greatly appreciate it.

Regards,
Shobana

Re: [BProc] swapping kmonte with kexec

[Joshua A. <lu...@ln...>]

Attached are the somewhat cleaned up kexec in bproc patches.

On Tue, 2005-10-18 at 12:19 -0600, Joshua Aune wrote:
> I have some working test patches to beoboot that swaps out kmonte with
> kexec.  So far I have booted one node using it and would love some
> further testing.
> 
> Benefits to kexec:
>  more arch support (works with x86_64 on stage1!)
>  is now in the kernel
>  works with smp kernels
> 
> This has allowed me to use a single kernel build for both stage1 and
> stage2 images (yay!).
> 
> These patches depend on mkelfImage and a statically build kexec binary.
> 
> If you are interested, let me know.  I'd post them but they will
> probably be changing over the next little while as I start cleaning them
> up.
> 
> Josh
> 
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Power Architecture Resource Center: Free content, downloads, discussions,
> and more. http://solutions.newsforge.com/ibmarch.tmpl
> _______________________________________________
> BProc-users mailing list
> BPr...@li...
> https://lists.sourceforge.net/lists/listinfo/bproc-users

[BProc] swapping kmonte with kexec

[Joshua A. <lu...@ln...>]

I have some working test patches to beoboot that swaps out kmonte with
kexec.  So far I have booted one node using it and would love some
further testing.

Benefits to kexec:
 more arch support (works with x86_64 on stage1!)
 is now in the kernel
 works with smp kernels

This has allowed me to use a single kernel build for both stage1 and
stage2 images (yay!).

These patches depend on mkelfImage and a statically build kexec binary.

If you are interested, let me know.  I'd post them but they will
probably be changing over the next little while as I start cleaning them
up.

Josh

Re: [BProc] Patch to beoboot-cm1.10 for module deps

[Michal J. <mi...@ha...>]

On Thu, Oct 13, 2005 at 05:22:52PM -0600, Joshua Aune wrote:
> The attached patch adds a check and
> error message.

Is a variation "dependancy" <-> "dependency" in an error message
to distinguish between cases?  Pretty subtle. :-)

  Michal

[BProc] Patch to beoboot-cm1.10 for module deps

[Joshua A. <lu...@ln...>]

Just had a fun error where the stage 2 image couldn't resolve a module
dependancy and was seg faulting.  The attached patch adds a check and
error message.  Who do I submit patches to?

Thanks,
Josh

[BProc] /dev/zero (deleted)

[Julian S. <ju...@va...>]

I'm running Valgrind on a BProc cluster.  Every now and again 
V looks at /proc/self/maps to make sure it hasn't lost track of
the process' address space layout.

Mostly this is fine.  However, mappings which started out referring
to "/dev/zero" seem to get renamed to "/dev/zero (deleted)", possibly
as a result of process migration (not sure).  Also, the mapping's
device and inode numbers change.

This isn't harmful, but I am curious to know what's going on.
It doesn't happen to anonymous mappings, which is a bit strange.

J

Re: [BProc] processes timeout after 1 second

[Andrew P. <ap...@ro...>]

Greg,

Yes, using bjs, I've now added the (-s numberofseconds) argument and
that did it, thank you very much for your help!

- Andrew

On Sep 27, 2005, at 11:50 AM, Greg Watson wrote:

> Are you using bjs to allocate the nodes? The default allocation  
> time is 1 second I think.
>
> Greg
>
> On Sep 27, 2005, at 8:34 AM, Andrew Pitre wrote:
>
>
>> I'm having trouble getting mpi programs to execute for >= 1 sec.   
>> I have a simple program that loops, prints the execution time then  
>> quits.
>>
>> When the loop count is increased to where the execution time is  
>> greater than or about 1 sec, the program fails with the following  
>> messages:
>> "mpirun: error: child process (rank=0; node=0) exited abnormally.
>> mpirun: error: aborting."
>>
>> Replacing the loop with a sleep() statement has a similar effect,  
>> processes can sleep for any amount of time < 1 sec, e.g. sleep(. 
>> 999999) is ok, but if sleep(1) is called the program fails with  
>> the above error.
>>
>> I've tried adjusting the pingtimeout with settings 30, 3000, and  
>> 30000, without success.
>>
>> The environment is Clustermatic 5 with a custom compiled 2.6.9  
>> kernel and bproc4.0.0pre8 module on Opteron processors.  This  
>> problem does not appear on a LAM based non-bproc cluster with the  
>> same source code.
>>
>> Any help with this will be greatly appreciated.
>>
>> - Andrew
>>
>>
>>
>>
>>
>> -------------------------------------------------------
>> SF.Net email is sponsored by:
>> Tame your development challenges with Apache's Geronimo App  
>> Server.Download it for free - -and be entered to win a 42" plasma  
>> tv or your very
>> own Sony(tm)PSP.  Click here to play: http://sourceforge.net/ 
>> geronimo.php
>> _______________________________________________
>> BProc-users mailing list
>> BPr...@li...
>> https://lists.sourceforge.net/lists/listinfo/bproc-users
>>
>>
>
>

Re: [BProc] processes timeout after 1 second

[Greg W. <gw...@la...>]

Are you using bjs to allocate the nodes? The default allocation time  
is 1 second I think.

Greg

On Sep 27, 2005, at 8:34 AM, Andrew Pitre wrote:

> I'm having trouble getting mpi programs to execute for >= 1 sec.  I  
> have a simple program that loops, prints the execution time then  
> quits.
>
> When the loop count is increased to where the execution time is  
> greater than or about 1 sec, the program fails with the following  
> messages:
> "mpirun: error: child process (rank=0; node=0) exited abnormally.
> mpirun: error: aborting."
>
> Replacing the loop with a sleep() statement has a similar effect,  
> processes can sleep for any amount of time < 1 sec, e.g. sleep(. 
> 999999) is ok, but if sleep(1) is called the program fails with the  
> above error.
>
> I've tried adjusting the pingtimeout with settings 30, 3000, and  
> 30000, without success.
>
> The environment is Clustermatic 5 with a custom compiled 2.6.9  
> kernel and bproc4.0.0pre8 module on Opteron processors.  This  
> problem does not appear on a LAM based non-bproc cluster with the  
> same source code.
>
> Any help with this will be greatly appreciated.
>
> - Andrew
>
>
>
>
>
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App  
> Server.Download it for free - -and be entered to win a 42" plasma  
> tv or your very
> own Sony(tm)PSP.  Click here to play: http://sourceforge.net/ 
> geronimo.php
> _______________________________________________
> BProc-users mailing list
> BPr...@li...
> https://lists.sourceforge.net/lists/listinfo/bproc-users
>

[BProc] processes timeout after 1 second

[Andrew P. <ap...@ro...>]

I'm having trouble getting mpi programs to execute for >= 1 sec.  I  
have a simple program that loops, prints the execution time then quits.

When the loop count is increased to where the execution time is  
greater than or about 1 sec, the program fails with the following  
messages:
"mpirun: error: child process (rank=0; node=0) exited abnormally.
mpirun: error: aborting."

Replacing the loop with a sleep() statement has a similar effect,  
processes can sleep for any amount of time < 1 sec, e.g. sleep(. 
999999) is ok, but if sleep(1) is called the program fails with the  
above error.

I've tried adjusting the pingtimeout with settings 30, 3000, and  
30000, without success.

The environment is Clustermatic 5 with a custom compiled 2.6.9 kernel  
and bproc4.0.0pre8 module on Opteron processors.  This problem does  
not appear on a LAM based non-bproc cluster with the same source code.

Any help with this will be greatly appreciated.

- Andrew