I am new to HBASE and Phoenix world. I have designed and executed a map reduce job which writes around 2.4 billion cells (rows*columns) in HBASE via Phoenix in about 80min. I have reduced the "mapreduce.input.fileinputformat.split.maxsize" to 8MB to increase the number of mapper which helped me to increase throughput and reduce duration from 140min to 80min. I have wide rows i.e each row has about ~27-30k columns and most of these ( 95%) are dynamic columns which are created during UPSERT operations.
I wanted guidance with if I can reduce the time duration of my job any further by tweaking any parameters which are default right now from either HBASE side or Phoenix side. Thanks.
