1

I am trying to implement JCUDA code on Hadoop. However while running the application, it gives unsatisfied link error:

can not load the native library "JCudaDriver-linux-x86_64" with base name "JCudaDriver"'

I have tried added the path to the library in LIBJARS, LD_LIBRARY_PATH, etc. I have tried setting the variable in the Java code using System.setProperty(). I have also tried the variable -Djava.library.path. However nothing seems to be working.

Operating System: Linux
Jcuda version: 0.7.0
CUDA version: 7.0
Hadoop version: 1.2.1

JoshDM
  • 4,939
  • 7
  • 43
  • 72
Anikam
  • 17
  • 3
  • Thanks for the edit, Kevin! – Anikam May 22 '15 at 04:32
  • Hadoop spawns "child JVMs" on the nodes, and you might have to adjust your configuration with an **additional** setting of the `java.library.path` for the child VMs, as described in http://hadoop.apache.org/docs/r1.2.1/mapred_tutorial.html#Task+Execution+%26+Environment (at least, I know that people once got this working, according to http://forum.byte-welt.net/byte-welt-projekte-projects/swogl-jcuda-jocl/jcuda/3188-jcuda-hadoop-exception.html ...) – Marco13 May 26 '15 at 10:25
  • I'm curious, was the issue resolved by setting the additional `java.library.path` ? – Marco13 Jun 10 '15 at 15:49
  • Sorry for the late reply. Yes it was. Thanks for the help! – Anikam Jul 13 '15 at 05:08

0 Answers0