> > Hi hammer, > Thank you. I don't know Zebra, but it sounds very good.
> > > 发件人： Jeff Hammerbacher <ham...@cloudera.com> > 日期： 2010年2月5日 上午4:33 > 主题： Re: Use DTS instead of DFS for data warehouse > 收件人： hdfs-dev@hadoop.apache.org > > > > Hey 易剑, > > Your proposed system sounds quite a bit like Zebra, which is a contributed > project under the Pig subproject: http://wiki.apache.org/pig/zebra. Have > you > taken a look at Zebra? > > Thanks, > Jeff > > 2010/2/4 易剑 <myhad...@gmail.com> > > > *Glossary* > > DTS: Distributed Table System, not a bigtable > > DFS: Distributed File System > > > > > > DFS is better for unstructed data, but DTS is better for structed data, > > data > > warehouse is structed, so I think a table is better than a file. DTS is > > following: > > 1. Break a logic big table into a many physical small table > > 2. The same size blocks is not necessary > > 3. The order of blocks is not necessary > > 4. Only store structed data > > 5. Support block indexes > > 6. Support deleting and updating > > 7. The interfaces are SQL, but only a block > > 8. Spliting a table horizontally and vertically is supported at the same > > time > > 9. 。。。 > > > >
