Just wondering, is there a reason why twitter doesn't use one of the many distributed in-memory database solutions?
It seems like they had to write a lot of custom layering on top of it just to scale
At a certain point of complexity and scale the in-house, custom distribution layer is almost always going to outperform a general purpose distribution system built into the database.
General purpose distributed database clusters are progressing, and if Riak, or one of the other in memory cluster focused system had been stable and production ready when Twitter was developing their cache layer it might have been a strong contender.
However, Riak is still much, much slower than Redis, especially when it comes to accepting writes. Overall, when you have the money that Twitter does and the team that it has you can come up with something better in house that it is more efficient for your use case. And that's what they've done here by building on top of Redis.
