Facebook messaging and choice of HBase over Cassandra - what can we learn?

You can learn that the decision was very complex with a lot of benefits and detriments for both stores. In the end, we thought that HBase gave us more initial benefits, reduced time to market, and a community with similar long-term goals.

http://www.facebook.com/notes/facebook-engineering/the-under... talks a bit more about this decision.

Funny, I (very grudgingly) scrapped our (upcoming) Cassandra deployment last week. It's a nice project, but there were too many bugs and glitches for our production requirements.

HBase is an interesting choice, although I'd need to see stronger public "trials by fire" before I'd use it - but then again, we don't have dev resources to spare (like Facebook does).

Interesting that they chose hbase (not theirs; a bigtable clone) over cassandra ("open sourced by Facebook in 2008").

Good for them, choosing the better tech over the in-house tech.

Did facebook drop Cassandra completely? Or do they just not use it for the messaging part?