Hi, I'm new to HBase. I'm working with Hadoop-1.0.3 and HBase 0.92.1. I have 8 data nodes which also work as region servers. And I'm trying to import my data into HBase. I wrote two programs, one is using HBase client API(auto flush is off, WAL is on, multi-threaded) and the other is using HBase MR like the Uploader example. Row key is a combination of several meaningful fields and a hash in the end, about 40 bytes or so. Each row is consist of several columns. There is only 2 column families. Average raw data of a row is less than 200 bytes. The one using client API is about 10,000 rows written per second, while the MR one is extremely slow, about 4~5 times slower. This is measured after a major compaction. Why the writing speed is so slow? Can someone help me.
-- language: Chinese, Japanese, English