10 Must know things about HBase performance

Facts your should know about Hbase

1. Write in Hbase takes  - 1-3 ms

2. Read takes 0-3ms if data is cached, 10-30ms if not cached and Hbase reads it from disk.

3. Its preffered to work with 3mb cell sizes

4. Define region max size as bigger as possible to avoid split. Controlled by property hbase.hregion.max.filesize 

6. You can cancel major compaction by hbase.hregion.majorcompaction=0 and run in manually.

7. Know to define cache size of the client by hbase.client.scanner.caching. and setBatch  for  columns limitation.

8. Know to limit Map reduce task:
mapred.tasktracker.map.tasks.maximum and  mapred.tasktracker.reduce.tasks.maximum.

9. Use property hfile.block.cache.size  to define percent of your heap for block caching. Default - 0.25. Im using 0.5.

10. Enable Data copression - Use snappy or LZO.