Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
5.0.0
-
None
Description
I am logging this MB for tracking purposes (say in the future) we can consider an additional dimension to Multi-Dimensional Scaling (MDS) [Resource Pooling]
Today our MDS implementation requires parity in node capacity within the Services Types - Index, Query and Data.
Though you can deploy mixed node sizes there is a single Memory / Core Configuration for all node types - Index , Data, Query. Deploying disparate machine types just skews the usage of resources across the cluster.
There is no disagreement that Data node types need to be 100% identical since they are hash partitioned. Query and Indexes however are stateless and non-partitioned services, there are advantages/benefits in having disparity in the instance types e.g. with resource pooling
e.g.
- High Capacity Index nodes with Larger Memory, Larger Core Counts (Would not require index partitioning)
- Higher or Lower Core / Memory Query Services can be used to cater to other use cases without interrupting production usage
Note: from a customer standpoint this makes a significant impact on the TCO and ease of use conversation. e.g. I don't have to have all i2 Instances which are exorbitantly expensive