PVM errors at startup

James Kohl kohlja at ornl.gov
Tue Oct 14 15:08:54 EDT 2003


Hey Patrick,

Glad you found the problem.  This is usually manifested when the
networking config is off slightly, or when internal/external networks
are confused, but it sounds like you had a much more interesting
problem...!  :-)

Yes, PVM uses rsh/ssh/TCP to start a remote PVM daemon (pvmd) but
then the daemons themselves use UDP to talk and route PVM messages.
FYI, any PVM tasks that use the "PvmRouteDirect" will use direct TCP
sockets.

Again, glad you figured it out!  (And you're most welcome! :)

All the Best,

	Jim

  On Tue, Oct 14, 2003 at 06:53:54PM +0200, Patrick Begou wrote:
  > This email just to close the thread with the solution.

  > The problem was not related to any PVM misconfiguration but to the
  > ethernet driver. Looking at the ethernet communications between 2 nodes
  > with tcpdump has shown that pvmd was started using tcp communications
  > BUT that pvmd were trying to talk each other with UDP protocol (it is
  > also detailed in the PVM doc) and this was the problem. The UDP
  > communications was unsuccessfull between the nodes.

  > Details:
  > The nodes are P4 2.8 with Asustek P4P800 motherboard and on board 3C940
  > (gigabit) controler. I was using the 3c2000 driver (from the cdrom).
  > Kernel is 2.4.20-20.7bigmem from RedHat 7.3.
  > rsh, rexec and rcp are working fine but this driver seems not to work
  > with UDP protocol???

  > The solution was to download the sk68lin driver (v6.18) and run the
  > shell script to patch the kernel sources for the current kernel. Then
  > correct the module.conf file and set up the gigabit interface. Now PVM
  > is working fine between the two first nodes and the measured throughput
  > is the same as with 3c2000 asustek driver. I should now setup the other
  > nodes!

  > I would like to thanks Pr. Kenneth R. Koehler and Dr James Arthur Kohl
  > for their great help in checking the full PVM configuration and leading
  > me towards a network driver problem.

  > Patrick

(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:(:

   James Arthur "Jeeembo" Kohl, Ph.D.     "Da Blooos Brathas?!  They
   Oak Ridge National Laboratory              still owe you money, Fool!"
   kohlja at ornl.gov
   http://www.csm.ornl.gov/~kohl/          Long Live Curtis Blues!!!

:):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):):)
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list