Spark Hydra This program is an application for performing proteomic search on a spark cluster. The program contains multiple search algorithms but most work has been done on a version of comet. The code depends on another GitHub Project - Spark Accumulators primarily to provide profiling support. The current version of the code depends on a Spark 1.5 cluster. The command to run depends on two files - one describing the cluster and a second describing the proteomic search independent of the cluster. The command line looks line the following: spark-submit --class com.lordjoe.distributed.hydra.comet_spark.SparkCometScanScorer SteveSpark.jar ~/SparkClusterEupaG.properties input_searchGUI_scan10000.xml Where spark-submit - standard spark command --class com.lordjoe.distributed.hydra.comet_spark.SparkCometScanScorer define the main class HydraSpark.jar define the jar file ~/SparkClusterEupaG.properties file decsribing the cluster input_searchGUI_scan10000.xml file describing the search job

Cluster descriptive file The location and access to the cluster are described in the default config cluster. This file sets a series of spark properties. Properties starting with com.lordjoe are hydra specific ‘A couple of properties of special interest are com.lordjoe.distributed.PathPrepend=hdfs://daas/steve/eg3/ PathPrepend is a string prepended to file names to reference the data in hdfs. Because the files for a problem are frequently placed in a directory which is mapped or copied to hdfs this text allows the file describing various files such as spectra and database files to be described without reference to the hdfs implementation. Thus a file such as Tandem.xml would be accessed by spark as hdfs://daas/steve/eg3/Tandem.xml .

These are properties to be set on the spark cluster

prepend to path

com.lordjoe.distributed.PathPrepend=hdfs://daas/steve/eg3/

spark.mesos.coarse=true spark.mesos.executor.memoryOverhead=3128

com.lordjoe.distributed.hydra.BypassScoring=true com.lordjoe.distributed.hydra.KeepBinStatistics=true com.lordjoe.distributed.hydra.doGCAfterBin=false

give executors more memory

spark.executor.memory=12g

Spark shuffle properties

spark.shuffle.spill=false spark.shuffle.memoryFraction=0.4 spark.shuffle.consolidateFiles=true spark.shuffle.file.buffer.kb=1024 spark.reducer.maxMbInFlight=128

spark.storage.memoryFraction=0.3 spark.shuffle.manager=sort spark.default.parallelism=360 spark.hadoop.validateOutputSpecs=false

#spark.rdd.compress=true #spark.shuffle.compress=true spark.shuffle.spill.compress=true spark.io.compression.codec=lz4 spark.shuffle.sort.bypassMergeThreshold=100

try to divide the problem into this many partitions

com.lordjoe.distributed.number_partitions=360

Search descriptive file The location and access to the cluster are described in the default config cluster. This file follows the format of input for XTandem. list path, default parameters is the reference for a file holding common parameters not overtidden in the default file - note the path is absolute do the prepend string described above will not be added list path, taxonomy information is a file describing the database as specified by XTandem spectrum, path is a file containing spectra output, path is a file the resultant pep.xml Scoring, algorithm is either comet or xtandem

hdfs://daas/promec/eupa2015/parameters_searchGUI.xml taxonomy_searchGUI.xml uniprot_sprot_concatenated_target_decoy.fasta scan1000000.mzXML scan1000000.out.xml comet