ParBLiSS/parconnect

SIGSEGV

oehlrich9 opened this issue · 3 comments

oehlrich9 commented

On testing the parconnect for the SC16, we run into some segfault errors while using any type of data found in the original paper.
Where can we get some input files suitable for the application?
I attached the output for investigation what went wrong.

parconnectout.txt

patflick commented

Thank you for the report.
In order to reproduce your issue, could you elaborate on:

  • MPI Version
  • Compiler Version
  • How many MPI processes did you run this on?
  • Which input dataset are you using?
  • What were the exact parameters you started the application with?
  • Hardware (available memory, count of cores, CPU)

Thanks.

oehlrich9 commented

Hi

MPI Version:
intel/5.1up03-intel

Compiler Version:
gcc/5.3.0

How many MPI processes did you run this on?
40

Which input dataset are you using?
mgm4539518.3.050
Thats one of the original Datersets found with the project number from the paper

What were the exact parameters you started the application with?
-np 40 ./benchmark_parconnect_auto --input generic --pointerDouble y --file /elxfs/rzku/hpcv001h/mgm4539518.3.050.upload.bin

Hardware (available memory, count of cores, CPU)
I'm runnig the application on one node of our biggest university cluster:
(https://www.rrze.fau.de/dienste/arbeiten-rechnen/hpc/systeme/emmy-cluster.shtml)

2x 2.2GHz Skylake CPUs with 10 pysical, 20 logical cores each
64 GB RAM

Thanks for your help.

cjain7 commented

Hi oehlrich9,

The input type should be dbg here : --input dbg. generic is supposed to be used when you have graph input in the edgelist format.

Also, I hope your file mgm4539518.3.050.upload.bin is in the appropriate fastq format. I am just being skeptical because .bin is not the extension of these files usually (unless you named it so)

Let me know if you still face any problem.

PS: README file of this repository has the link to SC16 specific version of parconnect.