Capacity Planning

• How many nodes?
• The basic starting point is two nodes with 2 cores and 4GB of memory on each node
• For fault tolerance perspective three nodes is more appropriate for any cluster
• What's better more nodes or bigger nodes?
• More nodes equals IO, Memory, and GC (garbage collector) distributed processing
• Common pitfall with distributed databases - stressing common storage e.g. SAN (system attached storage)
• Bigger nodes means more processing can be performed on a node with fast access to in-memory data and faster local IO
• Resizing node in production is likely more challenging than adding a new node to the cluster
• Elasticsearch is built for scaling out on commodity hardware, not up on single massive machine
• How high can it go? Pretty high
• So which one it is going to be: more smaller nodes or less larger nodes?