[SPARK-1888] enhance MEMORY_AND_DISK mode by dropping blocks in parallel

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Component/s: Spark Core
• Assignee: Wenchen Fan
• Reporter: Wenchen Fan
• Created: 20/May/14 02:58
• Updated: 11/Dec/14 20:39
• Resolved: 11/Dec/14 20:39

Description

Sometimes MEMORY_AND_DISK mode is slower than DISK_ONLY mode because of the lock on IO operations(dropping blocks in memory store). As the TODO says, the solution is: only synchronize the selecting of to-be-dropped blocks and do the dropping in parallel. I have a quick fix in my PR: https://github.com/apache/spark/pull/791#issuecomment-43567924
It's fragile currently but I'm working on it to make it more robust.

Issue Links

• duplicates: SPARK-3000 Drop old blocks to disk in parallel when memory is not large enough for caching new blocks
• links to: [Github] Pull Request #791 (cloud-fan)