Shared resource pool with periodic rebalancing in a multi-core system

A data storage system comprising:

• processing circuitry and a memory;
• a plurality of non-volatile data storage drives;
• and
• wherein the memory has program code stored thereon, wherein the program code, when executed by the processing circuitry, causes the processing circuitry to:
• load a predetermined percentage of a plurality of resource units contained in a shared resource pool of a data storage system into a global partition within the shared resource pool;
• load those resource units contained in the shared resource pool that were not loaded into the global partition into a plurality of per-core partitions within the shared resource pool, wherein each one of the per-core partitions corresponds to a respective one of a plurality of processor cores in the data storage system;
• for each one of the per-core partitions, allocate resource units from that per-core partition only to a work flow executing on the corresponding processor core;
• and
• periodically rebalance each one of the per-core partitions by moving resource units between the global partition and the per-core partition, at least in part by causing the processing circuitry to:
• calculate a local exhaustion time interval for the per-core partition, wherein the local exhaustion interval comprises an estimated amount of time until all resource units currently available in the per-core partition are exhausted;
• obtain a target exhaustion time interval, and
• move resource units from the global partition to the per-core partition in response to a determination that the local exhaustion time interval for the per-core partition is smaller than the target exhaustion time interval;
• and
• wherein the processing circuitry is triggered to periodically rebalance at least one of the per-core partitions by a determination that an amount of time since a last time that rebalancing was performed on the per-core partition exceeds a maximum rebalance time interval.

A computer program product including a non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed on processing circuitry, cause the processing circuitry to perform steps including:

• loading a predetermined percentage of a plurality of resource units contained in a shared resource pool of a data storage system into a global partition within the shared resource pool;
• loading those resource units contained in the shared resource pool that were not loaded into the global partition into a plurality of per-core partitions within the shared resource pool, wherein each one of the per-core partitions corresponds to a respective one of a plurality of processor cores in the data storage system;
• for each one of the per-core partitions, allocating resource units from that per-core partition only to a work flow executing on the corresponding processor core;
• periodically rebalancing each one of the per-core partitions by moving resource units between the global partition and the per-core partition, wherein the rebalancing includes:
• calculating a local exhaustion time interval for the per-core partition, wherein the local exhaustion interval comprises an estimated amount of time until all resource units currently available in the per-core partition are exhausted,
• obtaining a target exhaustion time interval, and
• moving resource units from the global partition to the per-core partition in response to a determination that the local exhaustion time interval for the per-core partition is smaller than the target exhaustion time interval;
• and
• wherein periodically rebalancing at least one of the per-core partitions is triggered by a determination that an amount of time since a last time that rebalancing was performed on the per-core partition exceeds a maximum rebalance time interval.