Date: Thu, 13 Sep 2012 13:11:13 -0500
From: Ross Reedstrom <reedstrm@rice.edu>
To: pgsql-performance@postgresql.org
Subject: AppScale backend datastore (NoSQL again kind of)
Message-ID: <20120913181113.GB26745@rice.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.5.21 (2010-09-15)

Hey PostgreSQL speed demons -
At work, we're considering an AppScale deployment (that's the Google App Engine
roll-your-own http://appscale.cs.ucsb.edu/). It supports multiple technologies
to back the datastore part of the platform (HBase, Hypertable, MySQL Cluster,
Cassandra, Voldemort, MongoDB, MemcacheDB, Redis). Oddly enough, in performance
tests, the MySQL Cluster seems like the general winner (their tests, we haven't
done any yet)

So, my immediate thought was "How hard would it be to replace the MySQL
Cluster bit w/ PostgreSQL?" I'm thinking hot standby/streaming rep. May or may
not need a pooling solution in there as well (I need to look at the AppScale
abstraction code, it may already be doing the pooling/direction bit.)

Any thoughts from those with more experience using/building PostgreSQL clusters?
Replacing MySQL Cluster? Clearly they must be using a subset of functionality,
since they support so many different backend stores. I'll probably have to
set up an instance of all this, run some example apps, and see what's actually
stored to get a handle on it. The GAE api for the datastore is sort of a ORM,
w/ yet another query language, that seems to map to SQL better than to NoSQL,
in any case. There seems to be a fairly explicit exposure of a table==class
sort of mapping.

Ross
-- 
Ross Reedstrom, Ph.D.                                 reedstrm@rice.edu
Systems Engineer & Admin, Research Scientist        phone: 713-348-6166
Connexions                  http://cnx.org            fax: 713-348-3665
Rice University MS-375, Houston, TX 77005
GPG Key fingerprint = F023 82C8 9B0E 2CC6 0D8E  F888 D3AE 810E 88F0 BEDE