[ https://issues.apache.org/jira/browse/MAPREDUCE-4502?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Tsuyoshi OZAWA updated MAPREDUCE-4502: -------------------------------------- Attachment: MAPREDUCE-4525-pof.diff I've prototyped proof of concept code of node-level aggregation. I made sure that it works well, so I share it! Current situation is as follows: * Node level aggregation in distributed mode is supported. I made sure of the correctness of the result by using diff command. * Fault tolerance is partially supported. At design level, it can be supported(WIP). * No test codes. * No switch on/off feature. Quick benchmark is as follows: |Default Hadoop(sec)|Node level aggregation(sec)| |105.0|82.0| Experiment Environment is as follows: * Input file is 1GB, generated by RandomTextWriter. * 8 CPU/machines. * 2 machines. I'm appreciate if you review the design and code. And, I have several questions to deal with: 1. How can I write tests such a complex case? Any ideas or references? 2. I've extended mr_protos.proto. Is it acceptable? My next step is creating patch sets with tests and fully fault tolerance support to merge it into trunk. I'm also planning to benchmark of node-level aggregation. I'll show you here when I do it. Note that design note is outdated, so I'm updating it now. If you'd like to track changes my code changes, you can see it on [my github|https://github.com/oza/hadoop-common/tree/MAPREDUCE-4525-Node-level-aggregation]. > Multi-level aggregation with combining the result of maps per node/rack > ----------------------------------------------------------------------- > > Key: MAPREDUCE-4502 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-4502 > Project: Hadoop Map/Reduce > Issue Type: Improvement > Components: applicationmaster, mrv2 > Reporter: Tsuyoshi OZAWA > Assignee: Tsuyoshi OZAWA > Attachments: design_v2.pdf, MAPREDUCE-4525-pof.diff, > speculative_draft.pdf > > > The shuffle costs is expensive in Hadoop in spite of the existence of > combiner, because the scope of combining is limited within only one MapTask. > To solve this problem, it's a good way to aggregate the result of maps per > node/rack by launch combiner. > This JIRA is to implement the multi-level aggregation infrastructure, > including combining per container(MAPREDUCE-3902 is related), coordinating > containers by application master without breaking fault tolerance of jobs. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira