I tried bonding two giganet cards on redhat 8.0 before. It seems that
bonding was good for Fast Ethernet, but not for giganet card. Litter
performance improvement was noticed for bonding giganet card.
Xuehua Chen
On Mon, 2003-10-06 at 15:37, Pierre Valiron wrote:
> Another alternative is to upgrade to lam 7.0.2.
>
> Then you may lamboot several independant "lam universes" after setting
> accordingly the environment variables LAM_MPI_SESSION_PREFIX and/or
> LAM_MPI_SESSION_SUFFIX.
>
> Easy then to lamboot one universe using some host-eth0 hostfile pointing
> to your eth0 set of interfaces, and another lam universe using some
> host-eth1 hostfile pointing to your eth1 set.
>
> Setting up bonding is substancially more tricky and needs root privileges,
> and some tweaking with the switch(es). I am not sure either it would be
> very efficient for gigabit.
>
> The best choice between a faster bonded network and two independant
> gigabit networks may also depend on the communication pattern of your
> two applications.
>
> Hoping this may help also.
> Pierre.
>
>
> On Mon, 6 Oct 2003, Nihar Sanghvi wrote:
>
> >
> > On Mon, 6 Oct 2003, Dmitry Kovalsky wrote:
> >
> > - Hi there
> > -
> > - I have two nodes. I use gigabit to run in parallel. However the CPUs are
> > - loaded to ~70% only. I want to setup another interface eth1 and to let
> > - another job to be computed there. By this I want to load the CPU at 100%.
> > - However seems LAM (6.5.9 I'm using) cann't run on eth0 and eth1 interfaces in
> > - simultaneously. Or I'm wrong?
> > -
> >
> > This basically is not an issue related to LAM. LAM does not know about
> > the number of underlying network cards. All it needs is an IP address to
> > communicate.
> > Two processes in the same LAM universe may not be able to use
> > different network cards.
> >
> > You could do something like Channel Bonding or unusual network topologies,
> > on the lower level. This is beyond what LAM does.
> >
> > Hope this helps..
> >
> > Nihar
> >
> >
> > - Sincerely yours,
> > -
> > - Ph.D. Student Dmytro Kovalskyy
> > - Institute of Molecular Biology & Genetics
> > - 150 Akad. Zabolotnogo Street,
> > - Kiev-143, 03143
> > - UKRAINE
> > -
> > - E-mail: dikov_at_[hidden]
> > - Fax: +380 (44) 266-0759
> > - Tel.: +380 (44) 266-5589
> > -
> > -
> > -
> > - _______________________________________________
> > - This list is archived at http://www.lam-mpi.org/MailArchives/lam/
> > -
> >
> >
> > Powered by LAM/MPI...
> > ---------------------------------------
> > Nihar Sanghvi
> > LAM/MPI Team
> > Graduate Student (Indiana University)
> > http://www.lam-mpi.org
> > --------------------------------------
> >
> > _______________________________________________
> > This list is archived at http://www.lam-mpi.org/MailArchives/lam/
> >
--
Xuehua Chen <cxh_at_[hidden]>
|
