[HADOOP-2560] Processing multiple input splits per mapper task

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Assignee: Dhruba Borthakur
• Reporter: Runping Qi
• Created: 09/Jan/08 16:34
• Updated: 17/Jul/14 19:31
• Resolved: 17/Jul/14 19:31

Description

Currently, an input split contains a consecutive chunk of input file, which by default, corresponding to a DFS block.
This may lead to a large number of mapper tasks if the input data is large. This leads to the following problems:

1. Shuffling cost: since the framework has to move M * R map output segments to the nodes running reducers,
larger M means larger shuffling cost.

2. High JVM initialization overhead

3. Disk fragmentation: larger number of map output files means lower read throughput for accessing them.

Ideally, you want to keep the number of mappers to no more than 16 times the number of nodes in the cluster.
To achive that, we can increase the input split size. However, if a split span over more than one dfs block,
you lose the data locality scheduling benefits.

One way to address this problem is to combine multiple input blocks with the same rack into one split.
If in average we combine B blocks into one split, then we will reduce the number of mappers by a factor of B.
Since all the blocks for one mapper share a rack, thus we can benefit from rack-aware scheduling.

Thoughts?

Issue Links

• is blocked by: HADOOP-249 Improving Map -> Reduce performance and Task JVM reuse
• is duplicated by: HADOOP-4565 MultiFileInputSplit can use data locality information to create splits
• is related to: MAPREDUCE-93 Job Tracker should prefer input-splits from overloaded racks
• is related to: HADOOP-3293 When an input split spans cross block boundary, the split location should be the host having most of bytes on it.
• is related to: HADOOP-249 Improving Map -> Reduce performance and Task JVM reuse