Capacity for in memory storage

Probably a very simple question.

Lets say I have worked it out that a namespace should be 110 gigs of data. That’s including secondary indices and friends.

Then I decide I want a replication factor of 2. And I have 17 nodes, and I want it all in memory.

So (110 * 2)/17 = 12.94 gigs per node.

Inside the namespace {} stanza, do I set memory-size to 110 gigs, 220 gigs or 13 gigs ?


13, but not really. You need to provide room for things to go wrong and memory fragmentation. Assuming you’re comfortable with a 70% HWM you’d need 13/0.7=~ 19G for each node.

Hi Albot: 19 is still fine. I expected to do a bit of maths, but yeah my real question was if it’s per node or not. Thanks so much.

Based on the data model, first size the “cluster” which has n nodes, allow for overheads - such as high water mark etc. Then based on the instances you want to use, decide n. n = (Cluster sizing value / Instance capacity value) plus 1 - to allow for rolling upgrades. Plan for 1 node down at a minimum.