From fddcdf87c919cadc5069b9ee3fc45396a24e8f9c Mon Sep 17 00:00:00 2001 From: Mosharaf Chowdhury <mosharaf@mosharaf-ubuntu.(none)> Date: Thu, 16 Dec 2010 11:58:00 -0800 Subject: [PATCH] Added a small description of how ParallelLFS works. --- src/scala/spark/BasicLocalFileShuffle.scala | 2 +- src/scala/spark/ParallelLocalFileShuffle.scala | 8 +++++--- 2 files changed, 6 insertions(+), 4 deletions(-) diff --git a/src/scala/spark/BasicLocalFileShuffle.scala b/src/scala/spark/BasicLocalFileShuffle.scala index aa83e5cf8c..57e8abf964 100644 --- a/src/scala/spark/BasicLocalFileShuffle.scala +++ b/src/scala/spark/BasicLocalFileShuffle.scala @@ -9,7 +9,7 @@ import scala.collection.mutable.{ArrayBuffer, HashMap} /** - * A simple implementation of shuffle using local files served through HTTP. + * A basic implementation of shuffle using local files served through HTTP. * * TODO: Add support for compression when spark.compress is set to true. */ diff --git a/src/scala/spark/ParallelLocalFileShuffle.scala b/src/scala/spark/ParallelLocalFileShuffle.scala index af461e48c0..7c957aba42 100644 --- a/src/scala/spark/ParallelLocalFileShuffle.scala +++ b/src/scala/spark/ParallelLocalFileShuffle.scala @@ -10,7 +10,9 @@ import scala.collection.mutable.{ArrayBuffer, HashMap} /** - * A simple implementation of shuffle using local files served through HTTP. + * An implementation of shuffle using local files served through HTTP where + * receivers create simultaneous connections to multiple servers by setting the + * 'spark.parallelLocalFileShuffle.maxConnections' config option. * * TODO: Add support for compression when spark.compress is set to true. */ @@ -221,12 +223,12 @@ object ParallelLocalFileShuffle extends Logging { if (!initialized) { // Load config parameters MinKnockInterval_ = System.getProperty ( - "spark.parallelLocalFileShuffle.minKnockInterval", "1000").toInt + "spark.parallelLocalFileShuffle.minKnockInterval", "1000").toInt MaxKnockInterval_ = System.getProperty ( "spark.parallelLocalFileShuffle.maxKnockInterval", "5000").toInt MaxConnections_ = System.getProperty ( - "spark.parallelLocalFileShuffle.maxConnections", "4").toInt + "spark.parallelLocalFileShuffle.maxConnections", "4").toInt // TODO: localDir should be created by some mechanism common to Spark // so that it can be shared among shuffle, broadcast, etc -- GitLab