[speedplane]

Is it just me, or is this article talking about load balancing, not sharding. My understanding of "sharding" is to split up a database into groups, either by time or by some index key (e.g., A-C on one shard, D-G on another, etc.). This article seems to be about splitting up web traffic, not sharding.

[icebraining]

Sharding is about splitting up the data in groups; in this case, the idea is that the web nodes have some local state, reducing the need to hit the databases so much:

"If all the clickstream clicks from the same user or state-change events from the same workflow or whatever go to the same host, you can hold the relevant state in that host’s memory, which means you can respond faster and hit your database less hard."