Even after major compacting it all? hbase> major_compact "TABLENAME"
.. then wait a while.... or just leave the cluster up 24 hours and do your measurement again. What did you du? du /hbase or du /hbase/TABLENAME? The former will include all WAL logs still outstanding. If size is a concern, run w/ lzo. See wiki page for how-to. St.Ack On Tue, Mar 30, 2010 at 10:45 PM, Chen Bangzhong <bangzh...@gmail.com> wrote: > Hi, ALL > > I am benchmarking HBase. I found that HBase used much more space than actual > size. Here is my test environment. > > One NameNode Server > One JobTracker Server (Secondary NameNode also on this machine) > One DataNode > > dfs.replication set to 1 > > <property> > <name>dfs.replication</name> > <value>1</value> > </property> > > My HBase Cluster includes one Master, one region server and one zookeeper on > 3 servers. > > I used the example code in HBase documentation to fill the test table. From > hadoop, I found that the space used is about 3 times the actual size. > > for example, I wrote 10k records to the table, each record is about 20k, the > actual size would be 2G. But from hadoop du command, the size used is more > than 6G. > > I don't know if this is by design? Or my configuration is wrong. > > thanks >
