Hi, I am wondering if when transfer happens between nodes due to shuffle data or hdfs blocks, if multiple threads act simultaneusly or not on the sending node. For example, on the reduce side, multiple fetchers will start each in a different thread. But on the mapper machine, for example if 5 fetchers comes for a job X, will the requests handled by ShuffleHandler, be treated in parallel, or one by one ?
Also, is this also the case with HDFS(BlockSender) ? Thanks in advance, Robert
