Re: Hypertable removed in AppScale

14 views
Skip to first unread message

Navraj S. Chohan

unread,
Nov 25, 2014, 11:23:09 AM11/25/14
to Sebastian Stenzel, appscale_...@googlegroups.com
Hey Sebastian,
CC'ing the mailing list.

We removed all datastores except Cassandra for ease of maintenance. There is a multiple of testing involved as we add more datastores, not to mention having far more dependencies. 

I addressed why we choose Cassandra here:

You can bring back HBase and Hypertable if you look at the diffs from the pulls. If you're looking to compare datastores it makes sense, otherwise I would just stick with Cassandra. The biggest downside that comes to mind having Cassandra is that since we use ByteOrderedPartitioning we are in charge of doing the load balancing (we must assign which nodes are in charge of which keyspace: https://github.com/AppScale/appscale/pull/1615/files). HBase and Hypertable, because they have a master/slave architecture, are closer to the original BigTable implementation in this respect, as hot spots will split up and be distributed over different tablet servers.

Thanks
Raj

On Tue, Nov 25, 2014 at 6:04 AM, Sebastian Stenzel <sebastia...@gmail.com> wrote:
Hi Navraj,

about a year ago you removed some datastore implementations from the appscale installation scripts. Today the only possible software is Cassandra. As we’re currently evaluating our possibilities with AppScale in a university research project, I’d like to know what were the reasons for this step.

Also: Is it still possible to use Hypertable or HBase by restoring the relevant chunks from the git history? Or would you advise to stick with Cassandra?

Thanks and best regards
Sebastian



--
Navraj Chohan (Raj), PhD
Co-Founder of AppScale Systems
A Google Cloud Partner 

and a GigaOm Structure/LaunchPad "Best of the Best" Startups


Santa Barbara, CA 93101

Reply all
Reply to author
Forward
0 new messages