[mvapich-discuss] LMC modifications

Abhinav Vishnu vishnu at cse.ohio-state.edu
Fri May 11 15:54:12 EDT 2007


Hi Chevchenkovic,

MVAPICH multi-rail device currently does not support LMC. We are working
on supporting multi-rail with LMC for MVAPICH2-1.0, which is the next
release and should be available in a couple of weeks.

Thanks much,

:- Abhinav

* On May,3 Chevchenkovic Chevchenkovic<chevchenkovic at gmail.com> wrote :
> I am looking at some modifications to the existing stack with multirail 
> support.
> So can someone suggest changes so that LMC enabled IB networks could be 
> used?
> -chev
> 
> 
> On 5/11/07, Abhinav Vishnu <vishnu at cse.ohio-state.edu> wrote:
> >Hi Chevchenkovic,
> >
> >Thanks for using MVAPICH. I hope that you are using MVAPICH 0.9.9
> >version for your evaluation.
> >
> >For your problem, you may not need to hardcode any values for using
> >LIDs 13 and 19. With 0.9.9 version, MVAPICH supports multi-pathing
> >using LMC mechanism. For your purpose, adding the following environment
> >variable during the execution of your program:
> >
> >% mpirun_rsh -np 2 node1 node2 VIADEV_USE_LMC=1 ./prog
> >
> >More details for this parameter are present at the following link of
> >MVAPICH user guide:
> >
> >http://mvapich.cse.ohio-state.edu/support/mvapich_user_guide.html#x1-690009.1.6
> >
> >Details about MVAPICH multi-pathing are present at:
> >
> >http://mvapich.cse.ohio-state.edu/support/mvapich_user_guide.html#x1-240005.7
> >
> >Please let us know the outcome of your experimentation.
> >
> >Thanks,
> >
> >:- Abhinav
> >
> >* On May,1 Chevchenkovic Chevchenkovic<chevchenkovic at gmail.com> wrote :
> >> Hi,
> >> I had this problem. I had the following configuration problem:
> >> node 1 : port 1  : LMC = 1 , LIDs = 12,13 (rank 0)
> >> node 2 : port 1  : LMC = 1 , LIDs = 18,19 (rank 1)
> >>
> >> I would like to perform send/recv using lid 13 and 19. I would like
> >> to hardcode these values. How do i proceed with the modifications in
> >> the MVAPICH stack?
> >> I dont mind hardcoding these details in the stack itself.
> >> Awaiting reply,
> >> -Chev
> >> _______________________________________________
> >> mvapich-discuss mailing list
> >> mvapich-discuss at cse.ohio-state.edu
> >> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
> >


More information about the mvapich-discuss mailing list