(Think I only replied to David, so forward it here...Having both asc/desc ordered CQ or RowKey is one way to solve it, and compression might reduce the disk being used, but I guess it still need to use double the amount of memory in cache.)
One cool feature in HBase 0.98 that Accumulo doesn't have currently is reverse scan. https://issues.apache.org/jira/browse/HBASE-4811 I already found a use case in my prototype! Jianshi On Wed, Jun 25, 2014 at 2:04 AM, Jianshi Huang <[email protected]> wrote: > Thank you David! Will do. > > > On Wed, Jun 25, 2014 at 2:02 AM, David Medinets <[email protected]> > wrote: > >> Each row is stored on one tablet server. If you have billions of >> entries with the same row then you can overload the tablet server's >> memory. Please read "Using Shards To Split a Row" at >> >> https://github.com/medined/D4M_Schema/blob/master/docs/data_distribution.md >> >> On Tue, Jun 24, 2014 at 1:59 PM, Jianshi Huang <[email protected]> >> wrote: >> > >> > On Wed, Jun 25, 2014 at 1:47 AM, David Medinets < >> [email protected]> >> > wrote: >> >> >> >> You're sharding the row value I hope? >> > >> > >> > Er... sorry, what do you mean? Some sort of reverted indexing? >> > >> > I might need to do it, right, then CF size will be quite large :), but >> I can >> > use an independent table, right? >> > >> > -- >> > Jianshi Huang >> > >> > LinkedIn: jianshi >> > Twitter: @jshuang >> > Github & Blog: http://huangjs.github.com/ >> > > > > -- > Jianshi Huang > > LinkedIn: jianshi > Twitter: @jshuang > Github & Blog: http://huangjs.github.com/ > -- Jianshi Huang LinkedIn: jianshi Twitter: @jshuang Github & Blog: http://huangjs.github.com/ -- Jianshi Huang LinkedIn: jianshi Twitter: @jshuang Github & Blog: http://huangjs.github.com/
