Yes - it could be separated at the cost of network io and data locality.
See this: http://www.larsgeorge.com/2010/05/hbase-file-locality-in-hdfs.html
On Tue, Nov 8, 2011 at 6:49 PM, edward choi <[EMAIL PROTECTED]> wrote:
> Pardon me for asking such a stupid question.
> I recently read an article about HBase basic architecture here:
> What really makes me wonder is, what if HBase is installed on machines
> where there is no HDFS?
> For example, there are ten linux machines (linux01, linux02, ... linux10),
> on which Hadoop 0.20.2 is installed, linux01 being Namenode and Jobtracker.
> Now, suppose that I install HBase on new computers linux11, linux12,
> linux13. Linux11 is the master and the rest are regionservers (Let's
> forget about zookeepers for argument's sake).
> If I configure hbase-site.xml so that hbase.rootdir would point to
> linux01, would this Hbase work properly?
> According to the article above, Hbase seems to have DFS clients of its
> own. So I thought HDFS and RegionServers could be separated. Am I
> right on this?