Forum: Managing VoltDB

Post: Server Sizing

Server Sizing
steves
May 1, 2011
Since much of the data in my application will be partitioned, presumably I don't need to size the RAM for each node to be able to handle the entire data set living on the cluster.

If I load a snapshot that contains all of the tables + data, does VoltDB require a box with enough RAM to handle the entire data set? Or does VoltDB immediately start moving data out to the nodes thereby lowering the requirement for a single monster server in the cluster dedicated to loading snapshots?

Thanks.
Steve.
SnapshotRestore is partition-aware
sebc
May 1, 2011
Steve,
No need for that "monster": the restore process is partition-aware.
That is indeed one great way to upgrade your cluster when you reach capacity.

Suppose you're at K=1 with 4 nodes (so each node has 1/2 the total data). You get at 80% memory and start worrying: build two new servers, snapshot the database, restart the cluster with 6 nodes and restore the database => each node now contains 1/3 of the total data and uses 54% of its memory.

Cheers,
Seb
Server Sizing
steves
May 3, 2011
That is truly elegant.
Thanks!