Data Load Balancing
NCache optimizes data distribution in Partitioned and Partition-Replica cluster caches using a distribution map based on a hash code of cache keys. This ensures data is almost evenly distributed across all nodes. Load balancing is critical for preventing performance bottlenecks caused by uneven key patterns or 'hot nodes' in a dynamic cluster.
Each node knows about the data load of every other node in the cluster. There can be scenarios where the data load on any one node becomes highly dependent upon the pattern of cache keys inserted in the cache.
NCache provides the following load balancing options:
Configuring Automatic Data Load Balancing
Configurable: Auto-load balancing is a configurable feature that automatically balances data load on every server node.
Automatic: It continuously monitors data load on all servers of the cluster after configurable intervals and detects uneven data load.
Threshold: It verifies the load against a configurable threshold, if any of the cache nodes have imbalanced data, then the coordinator generates new distribution maps to re-distribute the extra data from that node.
Turning Off: This feature can be turned off if the user does not want load balancing among cluster nodes.
Manual Rebalancing via NCache Management Center
Manual: Data load can be balanced at any time by the manual load balancing feature through the NCache Management Center.
No Threshold: There is no threshold. Data balancing can be started, whenever required.
See Also
Cache Serialization Format
Cache Server Backward Compatibility
Client Backward Compatibility
