You might want to have a look at this: 
http://www.larsgeorge.com/2009/10/hbase-architecture-101-storage.html

Insertions into HBase go in to memory first (into the MemStore) and only get 
flushed to disk periodically. When you do insertions sequentially in key order, 
your insert process will always hit one region at a time, which is handled by 
one region server. As such, it will give poor performance, because most of the 
region servers will be idling while only one is doing the work. It's better 
when insertions are distributed evenly across the key space, so all region 
servers get an equal share of work to do at the same time.



On 29 nov 2010, at 13:30, rajgopalv wrote:


Hi All,
I'm new to HBASE. I understand that HBASE keeps its data sorted in the
filesystem. So when we insert randomly, it takes time to sort. Where as when
we insert sequentially, there is no need for HBASE to sort.

But, i keep hearing from some of the users that, sequential inserts to HBASE
is the worst case thing. Why is that ?
--
View this message in context: 
http://old.nabble.com/Sequential-Inserts-In-HBASE.-tp30329923p30329923.html
Sent from the HBase User mailing list archive at Nabble.com<http://Nabble.com>.


Reply via email to