I think Steve forgot to reply to user@hbase. On Mon, Jul 29, 2013 at 10:11 AM, Steve Loughran <steve.lough...@gmail.com>wrote:
> I'm somewhere on safari in Tanzania right now, so all I have to add is > > http://steveloughran.blogspot.com/2013/06/hoya-hbase-on-yarn.html > > YARN lets you ask for compute/RAM capacity near the nodes you request, so > if/when region servers can be given the regions for the data closest to > them, you will be able to request region server instances near the data. > > 0.1 had to viewed as a throwaway PoC to show it can be done and identify > areas where both YARN and hbase need to be improved. For YARN there's an > uber JIRA. For hbase, apart from the locality of RS on startup, I'd like > the ports/URLs of the web servers of every node published on ZK or > elsewhere, as we have to say port 0 for all service ports, otherwise oh may > clash with existing instances > > ---------- ---------- > From: Michael Segel <mse...@segel.com> > Date: Mon, Jul 29, 2013 at 7:30 AM > Subject: Re: Anyone playing/looking at Hoya? > To: user@hbase.apache.org > > > Ted, > > Thanks, but I already read that, which prompted me to ask the question > about Hoya because I am curious about HBase on a YARN cluster. > > To me they seem orthogonal because you want HBase on the HDFS/Compute > nodes for data locality. > > My understanding of YARN is that you dynamically build the cluster around > your M/R job. Or is that an over simplification? > > Thx > > -Mike > > On Jul 27, 2013, at 6:00 PM, Ted Yu <yuzhih...@gmail.com> wrote: > > > http://hortonworks.com/blog/introducing-hoya-hbase-on-yarn/ > > >