I have found Apples to Apples (fsync off in PgSQL) and doing key/value store in PgSQL, PgSQL performed on par with, or slightly faster than MongoDB, using my admittedly not fully baked KVPBench app - http://github.com/gmr/kvpbench.
FUN FACT: (and you probably already know this but many do not): "fsync = off" is postgresql-ese for "please destroy all my data". Don't do it. Ever. The 9.0 docs are finally improving the language here to indicate that you must not set fsync off if you value your job.
Setting "synchronous_commit = off" nets you 99.9% of the performance improvement without the possibility of corrupting your entire database.
Where was the dba-running-with-scissors photo taken? Also, could we get the performance charts in your slides without the 3d-perspective? It's hard to tell relative performance from the angle.
Comments
I have found Apples to Apples (fsync off in PgSQL) and doing key/value store in PgSQL, PgSQL performed on par with, or slightly faster than MongoDB, using my admittedly not fully baked KVPBench app - http://github.com/gmr/kvpbench.
FUN FACT: (and you probably already know this but many do not): "fsync = off" is postgresql-ese for "please destroy all my data". Don't do it. Ever. The 9.0 docs are finally improving the language here to indicate that you must not set fsync off if you value your job.
Setting "synchronous_commit = off" nets you 99.9% of the performance improvement without the possibility of corrupting your entire database.
You might enjoy my fsync related photo:
http://www.flickr.com/photos/gavinmroy/4638958958/
And the presentation it came from:
http://www.scribd.com/doc/31669670/PostgreSQL-and-NoSQL
Very cool presentation.
why don't you publish the results of your benchmark?
I've been trying to find time to do a more in-depth test with multiple servers and include the scale out part of each system. I plan too soon.
Where was the dba-running-with-scissors photo taken? Also, could we get the performance charts in your slides without the 3d-perspective? It's hard to tell relative performance from the angle.
Outside our office in New Hope, PA. I'll do better graphs when I release the higher concurrency stats.