Running with Java
Workflows that require the Java Virtual Machine will run into a couple of issues when running on Genepool.
Java Software installed on Genepool
There are a number of software packages installed on Genepool which make use of the Oracle JVM. This isn't an exhaustive list however and is being actively maintained.
- MEGAN - Metagenomics analysis and visualization tool
Memory Issues with the JVM
If you find that your job is running out of memory or the JVM can not be initialized, this is because the JVM, by default will try to take 1/4 of the memory on the node where it is running. If you have not requested that much for your job (by using h_vmem), the JVM will not have enough memory and will crash. To ensure that you are setting the maximum heap size to the appropriate value, please set the following environment variable:
export _JAVA_OPTIONS="-Xms1G -Xmx1G"
set _JAVA_OPTIONS="-Xms1G -Xmx1G"
The "-Xms" and "-Xmx" options tell the JVM the minimum and maximum heap size respectively. In the example above, it is set to 1G. These arguments only accept integral values, so to set the heap size to something smaller, you can use "M" for megabytes, e.g. "-Xmx256M". You should select somewhat less memory for "-Xmx" than hv_mem (or ram.c for a single slot job) to account for overhead for any other processes your script may run concurrently.
Please note that the capitalization here is important, "-xmx1g", "-Xmx1g" will not work.
For more information, please check out this resource.
JVM not found
If you get the following error message:
testing JVM in /jgi/tools ...
testing JVM in /jgi/tools/SUN/jdk/1.6.0_13/jre ...
No suitable Java Virtual Machine could be found on your system.
The version of the JVM must be at least 1.5.
Please define INSTALL4J_JAVA_HOME to point to a suitable JVM.
You can also try to delete the JVM cache file /global/homes/a/aclum/.install4j
Then the JVM is not being found in the /jgi/tools directory or other directories specified in the .intstall4j file when the job is running is running on the compute nodes. To avoid this issue, set the following environment variable in your qsub script