Hi Alexandros, using gdb i found out that the program spins in line 116 : “graphlab::mpi_tools::init(argc,argv);” The backtrace that is produced by gdb is the following:
(gdb) backtrace #0 0x00007ffff55ef983 in __epoll_wait_nocancel () at ../sysdeps/unix/syscall-template.S:81 #1 0x00007ffff7570e03 in epoll_dispatch (base=0x123e000, tv=<optimized out>) at /home/user/graph-release/deps/event/src/libevent/epoll.c:407 #2 0x00007ffff6e97787 in opal_event_base_loop () from /usr/lib/libmpi.so.1 #3 0x00007ffff6ebaa3e in opal_progress () from /usr/lib/libmpi.so.1 #4 0x00007ffff386eef5 in orte_rml_oob_send () from /usr/lib/openmpi/lib/openmpi/mca_rml_oob.so #5 0x00007ffff386f404 in orte_rml_oob_send_buffer () from /usr/lib/openmpi/lib/openmpi/mca_rml_oob.so #6 0x00007ffff6e8739a in orte_routed_base_register_sync () from /usr/lib/libmpi.so.1 #7 0x00007ffff3e7cbb7 in ?? () from /usr/lib/openmpi/lib/openmpi/mca_routed_binomial.so #8 0x00007ffff6e69a32 in orte_ess_base_app_setup () from /usr/lib/libmpi.so.1 #9 0x00007ffff3a73ba6 in ?? () from /usr/lib/openmpi/lib/openmpi/mca_ess_singleton.so #10 0x00007ffff6e49843 in orte_init () from /usr/lib/libmpi.so.1 #11 0x00007ffff6e09024 in ompi_mpi_init () from /usr/lib/libmpi.so.1 #12 0x00007ffff6e2088f in PMPI_Init_thread () from /usr/lib/libmpi.so.1 #13 0x0000000000498f05 in init (argv=@0x7fffffffc310: 0x7fffffffe418, argc=@0x7fffffffc31c: 4) at /home/user/graph-release/src/graphlab/util/mpi_tools.hpp:59 #14 main (argc=4, argv=0x7fffffffe418) at /home/user/graph-release/toolkits/graph_analytics/tunkrank.cpp:116 I guess there is something wrong with MPI. Do you have any ideas? I have installed the packages openmpi-bin , openmpi-doc, libopenmpi-dev. Could anything else be missing too? Thanks again, Andreas On Jul 1, 2014, at 5:02 PM, Alexandros Daglis <[email protected]> wrote: > Dear Andrea, > > I have never encountered the problem you report for graph processing. Is > there any information being printed while it runs? > Also, have you tried using a debugger to identify where the application spins? > > Regards, > Alexandros > > On Jun 28, 2014, at 8:28 PM, Andreas Georgiadis <[email protected]> > wrote: > >> Hello, >> >> i am trying to run the web search, web serving and graph analytics >> benchmarks in my Linux VM, using Ubuntu 14.04 and i encounter the following >> problems: >> >> In the web search benchmark, the hadoop cluster does not start normally. >> When the datanode tries to connect to the port 9000 of localhost it >> terminates with an IO Exception, with the explanation ‘connection reset by >> peer’. >> >> In the graph analytics benchmark, when i run the tunkrank executable, it >> runs forever, even if i run it with the following configuration: >> >> ./tunkrank —powerlaw=10--ncpus=2 --engine=asynchronous >> >> Since the input graph has got only 10 vertices, i expected that the >> execution time would be short, so I suspect there is something wrong with >> the executable. >> >> Finally, in the web search benchmark, i get the following compilation errors >> when i execute make in the php-5.3.9 directory in the front end machine: >> >> /home/user/web-release/php-5.3.9/ext/dom/node.c: In function >> 'dom_canonicalization': >> /home/user/web-release/php-5.3.9/ext/dom/node.c:1898:21: error: >> dereferencing pointer to incomplete type >> ret = buf->buffer->use; >> ^ >> In file included from /home/user/web-release/php-5.3.9/main/php.h:38:0, >> from /home/user/web-release/php-5.3.9/ext/dom/node.c:26: >> /home/user/web-release/php-5.3.9/ext/dom/node.c:1900:40: error: >> dereferencing pointer to incomplete type >> RETVAL_STRINGL((char *) buf->buffer->content, ret, 1); >> ^ >> >> My gcc version is 4.8.2. >> >> Any piece of advice regarding those issues, in case you have encountered >> similar errors would help me a lot. >> >> I thank you in advance, >> Andreas Georgiadis >
