[Simgrid-user] SURF: Using GTNetS and network action delays

Kurt Vanmechelen kurt.vanmechelen at ua.ac.be
Wed Jul 16 11:09:07 CEST 2008


Hello again,

I recently tried to fire up SURF on GTNetS with a small test program but 
have run into the following error:
node -1 not found
** SimGrid: UNCAUGHT EXCEPTION received on (0): category: unknown_err; 
value: 0
** Cannot create GTNetS route
** Thrown by () in this process
[0.000000] xbt/ex.c:113: [xbt_ex/CRITICAL] Cannot create GTNetS route

**   In xbt_backtrace_current() at 
/home/kvanmech/workspace/simgrid/src/xbt/backtrace_linux.c:19
**   In route_onehop_new() at 
/home/kvanmech/workspace/simgrid/src/surf/network_gtnets.c:155
**   In surfxml_call_cb_functions() at 
/home/kvanmech/workspace/simgrid/src/surf/surfxml_parse.c:527
**   In ETag_surfxml_platform() at 
/home/kvanmech/workspace/simgrid/src/surf/surfxml_parse.c:262
**   In surf_parse_lex() at 
/home/kvanmech/workspace/simgrid/src/surf/simgrid_dtd.c:3766
**   In parse_platform_file() at 
/home/kvanmech/workspace/simgrid/src/surf/surfxml_parse.c:617

The problem seems to be located in the parse_platform_file call which 
builds up the GTNetS routes etc.
After a couple of calls to add_one_hop, the call to add_route returns -1 
with src == 1 and dst == 2. This is because
int dst_node = topo_->nodeid_from_hostid(dst); returns -1. That call 
returns 0 for the src_node btw.

I am testing this with the small_platform.xml from the repository, but 
the msg_platform.xml gave a similar error.

Here is my test code that leads up to the parse platform call:
cout << "GES_Surf test program!" << endl;
cout << "First argument (platform file) : " << argv[1] << endl;

surf_init(&args, argv);
surf_timer_model_init(argv[1]);

//surf_workstation_model_init_CLM03(argv[1]);
surf_cpu_model_init_Cas01(argv[1]);
surf_network_model_init_GTNETS(argv[1]);
surf_workstation_model_init_compound(argv[1]);

parse_platform_file(argv[1]);

Note that I can switch around network implementations such as Reno and 
Vegas just fine, only with GTNetS does the test program fail.

Am I doing something wrong here during initialization?

Thanks in advance for the feedback!

Best regards,

Kurt

Arnaud Legrand wrote:
> Hi,
>
> Kurt Vanmechelen wrote:
>
>> The GTNetS libraries were built on a Intel Quad Core Q9550 64bit 
>> using gcc 4.3. Due to the fact that 4.3 has cleaned up some header 
>> dependencies (see http://gcc.gnu.org/gcc-4.3/porting_to.html; C++ 
>> section), I had to add includes for e.g. <cstring> and <cstdlib> in 
>> the following files :
>> application-aodv.cc
>> application-sn.cc
>> application-webserver.cc
>> args.cc
>> memreuse.cc
>> nixvector.cc
>> routing-nixvector.cc
>> timerbuckets.cc
>
> If you still have the corresponding patch, could you send it to us. 
> That way, we could include it to the previous ones as a special 
> gcc-4.3 patch...
>
>> Aside from this the library built just fine.
>
> Cool.
>
> Cheers,
>
>     Arnaud
>





More information about the Simgrid-user mailing list