Partition View

BoostIO offers a distributed cache layer on the compute side. The BoostIO architecture design must consider not only service performance but also high availability indicators such as cache data reliability and cluster scalability. BoostIO uses the partition view solution to give full play to its distributed cluster system capabilities.

• The partition view is used for:
  • Copy management: Partitions are associated with copies. BoostIO supports dual copies, each of them is associated with two tiers of cache media, that is, memory and drives.
  • Data balancing: Based on the partition view and a load balancing algorithm, the cache client distributes service requests to each node in the cluster. This ensures balanced service load and cache utilization across the nodes to prevent single-point bottlenecks.
  • Linear scaling: After new nodes join the cluster, the cluster performance can be linearly increased by recalculating or changing the partition view.
  • Troubleshooting: The partition view can mark the cache status of each node and performs fault tolerance in case of faults to ensure service continuity.

• Working principle of the partition view:
  • The number of partitions remains unchanged after this configuration item is read during cluster initialization. An unchanged number prevents the global balancing policy from becoming invalid in scenarios such as capacity expansion and permanent node or process faults.
  • After the cluster is started, an initial partition view is generated based on the cluster node view, drive information of each node, and load balancing algorithm. This ensures that each node or drive carries an even number of partitions.
  • The partition view is recalculated when the cluster size grows or a permanent fault occurs. The recalculation process complies with data reliability, minimum data migration, and load balancing principles.