Forum: VoltDB Architecture

Post: Could one get resilience to partial failure with a cluster of clusters?

Could one get resilience to partial failure with a cluster of clusters?
monster
Oct 28, 2010
I'm ill today, and was daydreaming in bed...


I've just been wondering how difficult it would be to loosely connect several VoltDB clusters together. I'm thinking about applications that support several "independent logical units/locations", like forums, or solar systems in a space game. When one of the "units" fails, it should not affect the others, except that users cannot use the failed unit. A complete VoltDB cluster will halt if some of the data becomes (completely) unavailable, but not all data needs all other data to be useful. This is fine in many cases, but supporting partial failure through "cluster aggregation" sounds like a good feature as well. The independent clusters could be located in independent data-centers, to survive a data-center-wide breakdown. I would expect that it should not be too hard to achieve cross-cluster transactions, so that data can "move from one cluster to another" using some combination of "export" and Client API.


Has this concept already been discussed before?