This would be great, especially if there are in-memory and delicate map-reduce modes (see twister's task model http://www.iterativemapreduce.org/) to allow performance gains for small problems.
On Wed, Jun 9, 2010 at 5:03 PM, Jake Mannix <[email protected]> wrote: > Google paper <http://portal.acm.org/citation.cfm?id=1806596.1806638> > (requires > ACM membership) on their internal, higher-level java map/reduce api. > > Also of note, Jeff Hammerbacher seems to agree with me that this would be > *awesome* for Hadoop-land: > > https://issues.apache.org/jira/browse/MAPREDUCE-1849 > > Thoughts? > > I may just try to dredge up my initial pass at this kind of thing > (essentially a generified version of DistributedRowMatrix - DataSet<K,V>, > which then accepts Map<K,V,KOUT,VOUT> mappers and various reducers etc...), > and see if I can help with this. > > -jake >
