Most of the people I talk to with big sharded environments have had huge problems getting any sort of aggregation across the shards working correctly.

Timely replication of shared data to the individual shards can also be a huge issue.

Any architecture has it’s pros and cons. Normally the types of applications that work well in a sharded environment are ones that avoid the need for either of the above.

Unfortunately, many of the industry on-lookers (press etc) assume that any example of a successful architecture solving a given problem implies that that is the only possible solution for that problem. The reality is is that there are many ways to skin a cat.

Thank you for the clarification.

What is this “speed of light” problem you were referring to that applies to sharded databases but not to shared-nothing analytic database management?

And if that’s not your objection to shared-nothing analytic database management, then what — if anything — is? (You were quoted as having strong objection to same, but that quote was inaccurate, then I apologize for taking it literally. I’ve certainly been misquoted often enough … )

Best,

CAM