[MAPREDUCE-3788] [Gridmix] Investigate if Gridmix can be made YARN aware

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.0.0-alpha1
• Component/s: contrib/gridmix
• Labels: gridmix yarn
• Tags: gridmix yarn
• Assignee: Unassigned
• Reporter: Amar Kamat
• Created: 02/Feb/12 06:36
• Updated: 12/May/16 18:24

Description

Gridmix was written keeping in mind the monolithic JobTracker. Calls to the single JobTracker were throttled to avoid excess load. Also, polling was faster in JobTracker as the job statuses were cached even if the job was complete. In the YARN world, the situation is slightly different. To make Gridmix scalable and really a YARN scale-benchmarking tool, Gridmix should be enhanced. Some directions worth investigating are:
1. Investigate if Gridmix can cache the AM handles and poll the AM directly for map/reduce task progress.
2. Can the job monitor be made multi-threaded? Each thread can poll a bunch of AMs.
3. Check if there are better ways for getting job progress updates and get away with the busy-waiting logic in Gridmix.
4. Can Gridmix be made container aware. The definition of cluster load should be container aware.