BEAGLE + BEAST 2 in cluster 

Cluster provides more computational power. Paralleling your BEAST jobs and using SSE in cluster could save you more than half of your commutation time.

Here we show how BEAGLE options performed on the NeSI cluster (http://www.nesi.org.nz) and laptop (MacBookPro5,1: Intel Core 2 Duo). GPU option is not included in this comparison.

	Laptop	Cluster (Default)	Cluster OpenMP (4 threads)
Java	15m17s	11m55s	7m9s
BEAGLE	9m24s	7m45s	6m25s
BEAGLE SSE	8m1s	4m52s	3m50s

The speed is the minutes and seconds of per million samples, which is logged at the sample 36,000 by running the same xml using seed 777.

Command in laptop 

Use Java likelihood calculation only, BEAGLE is not considered:

    java -jar beast.jar -seed 777 -java my.xml 

Use beagle library but no SSE:

    java -jar beast.jar -seed 777 -beagle my.xml 

Use beagle library and SSE extensions:

    java -jar beast.jar -seed 777 -beagle_SSE my.xml

Default template for cluster 

The template below is to run a serial BEAST 2 job using BEAGLE SSE in cluster (Guide is here):

   #!/bin/sh
   #SBATCH -J FILE                 # The job name
   #SBATCH -A uoa???               # The account code
   #SBATCH --time=12:00:00         # The walltime
   #SBATCH --mem-per-cpu=16384     # Memory in MB ie. ? * 1024
   #SBATCH -D ./                   # The initial directory
   #SBATCH -o FILE_out.txt         # The output file
   #SBATCH -e FILE_err.txt         # The error file

   module load beagle-lib/20140322-goolf-1.5.14
   module load Java/1.8.0_5
   srun java -Xmx10g -Djava.library.path=$BEAGLE_LIB_PATH -jar beast.jar -beagle_SSE -seed 777 my.xml 

OpenMP (4 threads) template for cluster 

The 4-thread OpenMP (Open Multi-Processing) template below is to run a parallel BEAST 2 job using BEAGLE SSE in cluster (Guide is here):

   #!/bin/sh
   #SBATCH -J FILE                 # The job name
   #SBATCH -A uoa???               # The account code
   #SBATCH --time=12:00:00         # The walltime
   #SBATCH --mem-per-cpu=16384     # Memory in MB ie. ? * 1024
   #SBATCH --cpus-per-task=4       # 4 OpenMP Threads
   #SBATCH -D ./                   # The initial directory
   #SBATCH -o FILE_out.txt         # The output file
   #SBATCH -e FILE_err.txt         # The error file

   module load beagle-lib/20140322-goolf-1.5.14
   module load Java/1.8.0_5
   srun java -Xmx10g -Djava.library.path=$BEAGLE_LIB_PATH -jar beast.jar -threads 4 -beagle_SSE -seed 777 my.xml