Charm++ parallel programming system

[gaurav khanna <khannag AT cse.ohio-state.edu>]

Thanks Eric for the reply. I will try to figure out how to make changes
in the code based on your pointers.

Another question I have is regarding the mapping of processes to
processors. The AMPI manual does not talk about how to specify your own
mappings of processes onto processors as opposed to using the already
defined schemes like GreedyLB, TopoLB etc. Can you please provide inputs
with that?

Regards
Gaurav

Gaurav Khanna
Phd Student
CSE Department,OSU

Phone (office):614-292-7036

On Tue, 13 Feb 2007, Eric Bohm wrote:

> You should edit your netconfig file and set  USE_TRANSCEIVER 0.   That
> way it will use your trace files.  With USE_TRANSCEIVER set to 1, the
> simulator expects command line parameters directing it in which traffic
> patterns you wish it to generate.
>
>
> Link load can be obtained by setting DISPLAY_LINK_STATS to 1 as can
> message delay by DISPLAY_MESSAGE_DELAY 1.    The resolution of this data
> collection is controlled by the frequency set in COLLECTION_INTERVAL.
>
>
> Look in trunk/main/Nic.C NetInterface::recvPacket(Packet *p)  ~ line 400
> for an example.   Grep on config.collection_interval and you'll pretty
> much hit all the code sections involved in message and link statistics
> collection.
>
>
>
> gaurav khanna wrote:
> > Hello all,
> >
> > I am able to go one step furthur by being able to compile and run
> > charm++/AMPI on i686 machines. Previously when I had posted the query I
> > was trying it on x86_64 machines and it does not compile there.
> >
> > However, I am not still not able to successfully run BigNetSim. The readme
> > file README.bigsim explains the usage of BigNetSim using the traces
> > obtained by running BigSim. However, I am getting the following errors.
> >
> > I am using the program jacobi in the directory
> > charm/net-linux-bigemulator/examples/ampi/Cjacobi3D in my installation.
> >
> > I first run jacobi as follows.
> > ./charmrun +p2 jacobi +vp64 +x4 +y4 +z4 +cth1 +wth1 +bglog
> > This runs fine and generates trace files of the form bglog*. Now, I feed
> > it to a BlueGene Interconnection topology according to the instructions
> > in the README.bigsim file.
> > ./charmrun +p2 bigsimulator 1 0
> >
> > I get the following error.
> >
> >  Transceiver usage is pgm <trafficGenerator> <trafficPattern> <msgSize>
> > <loadfactor>
> > ------------- Processor 0 Exiting: Called CmiAbort ------------
> > Reason: Too few arguments
> > Stack Traceback:
> >   [0] CmiAbort+0x4a  [0x818169e]
> >   [1] _ZN6Config10readConfigEP8CkArgMsg+0x684  [0x810677e]
> >   [2] _ZN4mainC1EP8CkArgMsg+0x126  [0x80ea84c]
> >   [3] _ZN12CkIndex_main19_call_main_CkArgMsgEPvP4main+0x2b  [0x80eb421]
> >   [4] _Z10_initCharmiPPc+0xa31  [0x81431db]
> >   [5]
> > /j/osu3672/charm/net-linux-bigemulator/examples/ampi/Cjacobi3D/bigsimulator
> > [0x81847d4]
> >   [6] ConverseInit+0x2aa  [0x8184c19]
> >   [7] main+0x20  [0x814d7f4]
> >   [8] __libc_start_main+0xaa  [0x4012fbba]
> >   [9] gzprintf+0x51  [0x80ba411]
> > Fatal error on PE 0> Too few arguments
> >
> > Could you help me on this?
> >
> > Second thing as I have mentioned before is that my objective is to collect
> > performance characteristics like the total amount of communication on each
> > network link, # messages sent/recieved etc. I think your tool BigNetSim
> > looks very appropriate to do such a thing except that I will need to make
> > changes to the code to collect information like the #messages transferred
> > on each link, total data transferred on each link etc. If its not too much
> > of an effort for you, could you point me to places in code where this can
> > be done.
> >
> > Regards
> > Gaurav
> >
> > Gaurav Khanna
> > Phd Student
> > CSE Department,OSU
> >
> > Phone (office):614-292-7036
> >
> > On Wed, 7 Feb 2007, gaurav khanna wrote:
> >
> >
> >> Hi,
> >>
> >> I needed a simulator which allow me to simulate parallel codes on regular
> >> topologies like torus with an accurate model of interconnection network
> >> and contention. From my understanding, I think BigSim and BigNetSim is
> >> something that can help me do that.
> >>
> >> I tried to compile charm++ as follows.
> >>
> >> ./build bigsim mpi-linux bigsim -O
> >>
> >> I get the following error.
> >> Error> option: bigsim is not supported in this version!
> >> Supported compilers: cmpi cmpigcc cxx gcc gcc3 icc iccstatic kcc pgcc xlc
> >> xlc64
> >> Supported options: bluegene elan gm gm2 ifort lcs mpt ooc papi pgf90
> >> pthreads scyld smp vmi
> >>
> >> Does charm5.9 does not support BigSim? How can I fix this one.
> >>
> >> My objective is to collect performance characteristics like the total
> >> amount of communication on each network link, # messages sent/recieved
> >> etc. I think your tool BigNetSim looks very appropriate to do such a
> >> thing except that I will need to make changes to the code to allow me to
> >> collect communication performance characteristics. If its not too much
> >> of an effort for you, could you point me to places in code where this can
> >> be done.
> >>
> >> Regards
> >> Gaurav
> >>
> >> Gaurav Khanna
> >> Phd Student
> >> CSE Department,OSU
> >>
> >> Phone (office):614-292-7036
> >>
> >> _______________________________________________
> >> charm mailing list
> >> charm AT cs.uiuc.edu
> >> http://lists.cs.uiuc.edu/mailman/listinfo/charm
> >>
> >>
> >
> > _______________________________________________
> > charm mailing list
> > charm AT cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/charm
> >
>