Quality of service for internal I/Os using internal flow mechanism
A method for Quality of Service (QoS) for internal input/output (I/O) using an internal flow mechanism in a storage system is provided. The method includes establishing internal I/O flows corresponding to external I/O flows for one or more computing machines or virtual machines implemented with physical computing resources requesting I/Os. The method includes inheriting service-level agreement (SLA) values from the external I/O flows to the internal I/O flows and determining demands for the external I/O flows and the internal I/O flows. The method includes distributing credits to the internal I/O flows and the external I/O flows according to minimum I/O operations per unit time values, based on the service-level agreement values, and distributing remaining credits to the external I/O flows and the internal I/O flows according to priority and demand of each of the external I/O flows and the internal I/O flows.
1. A processor-based method for Quality of Service (QoS) for internal input/output (I/O) using an internal flow mechanism in a storage system, comprising:
establishing, in a storage system, internal I/O flows corresponding to external I/O flows for one or more computing machines or virtual machines implemented with physical computing resources requesting I/Os to the storage system;
inheriting service-level agreement (SLA) values from the external I/O flows to the internal I/O flows;
determining demands for the external I/O flows and the internal I/O flows;
distributing credits to the internal I/O flows and the external I/O flows according to minimum I/O operations per unit time values, based on the service-level agreement values; and
distributing remaining credits to the external I/O flows and the internal I/O flows according to priority and demand of each of the external I/O flows and the internal I/O flows.
2. The method of claim 1 , further comprising:
servicing in the storage system the internal I/O flows and the external I/O flows in accordance with the credits distributed thereto.
3. The method of claim 1 , wherein determining the demands is based on arrival I/O operations per unit time of a flow and based on total latency for serving and queuing the flow.
4. The method of claim 1 , wherein distributing the remaining credits according to the priority and demand comprises distributing credits proportional to priority of a flow.
5. The method of claim 1 , wherein distributing the remaining credits according to the priority and demand comprises, for at least one flow, distributing credits proportional to the priority for the flow and less than the demand for the flow.
6. The method of claim 1 , wherein the inheriting and the distributing remaining credits according to priority and demand result in an internal I/O flow corresponding to an external I/O flow of higher priority receiving more credits than an internal I/O flow corresponding to an external I/O flow of lower priority.
7. The method of claim 1 , wherein the demands are based on latency of underlying storage in the storage system and wherein distributing the credits based on the minimum I/O operations per unit time value and distributing the remaining credits according to the priority and demand ensure credits given to the external I/O flows are consistent with latency and throughput of the underlying storage.
8. A tangible, non-transitory, computer-readable media having instructions thereupon which, when executed by a processor, cause the processor to perform a method comprising:
tracking, in a storage system, for one or more computing machines or virtual machines implemented with physical computing resources, each of a plurality of external I/O (input/output) flows, having I/O requests for the storage system, and corresponding internal I/O flows;
inheriting service-level agreement (SLA) values including priorities and minimum I/O operations per unit time from the external I/O flows to the internal I/O flows;
calculating demands for each of the external I/O flows and each of the internal I/O flows;
assigning credits for each of the external I/O flows and each of the internal I/O flows, based on the minimum I/O operations per unit time in accordance with the service-level agreement values; and
assigning further credits for each of the external I/O flows and each of the internal I/O flows, based on the priorities and calculated demands.
9. The computer-readable media of claim 8 , wherein the method further comprises:
managing the internal I/O flows through at least one storage stack or storage queue in the storage system, on paths to underlying storage devices or storage tiers, consistent with the credits of the internal I/O flows.
10. The computer-readable media of claim 8 , wherein calculating the demands for each of the external I/O flows and each of the internal I/O flows comprises:
calculating a demand based on arrival I/O operations per unit time and total latency for serving and queuing, for each of the external I/O flows and each of the internal I/O flows.
11. The computer-readable media of claim 8 , wherein the assigning further credits for each of the external I/O flows and each of the internal I/O flows, based on the priorities and calculated demands comprises:
determining credits proportional to priority, for each of the external I/O flows and each of the internal I/O flows, based on the priorities from the service level agreements.
12. The computer-readable media of claim 8 , wherein the assigning further credits for each of the external I/O flows and each of the internal I/O flows, based on the priorities and calculated demands comprises:
determining credits proportional to priority and less than calculated demand, for each of the external I/O flows and each of the internal I/O flows.
13. The computer-readable media of claim 8 , wherein calculating the demands comprises:
setting a demand of a flow proportional to an arrival I/O operations per unit time of the flow if a total latency for serving and queuing the flow is less than or equal to a deadline for the flow based on the service-level agreement values and otherwise setting the demand of the flow proportional to a sum of the arrival I/O operations per unit time of the flow plus the total latency for serving and queuing the flow, divided by the deadline for the flow, for each of the external I/O flows and each of the internal I/O flows.
14. A storage system with an internal flow mechanism for Quality of Service, comprising:
at least one processor;
a first port configured to couple to one or more computing machines or virtual machines;
a second port configured to couple to a storage memory;
a flow tracker, configured to track external I/O (input/output) flows having I/O requests from the one or more computing machines or virtual machines and configured to track internal I/O flows corresponding to the external I/O flows;
the flow tracker configured to inherit, from the external I/O flows to the internal I/O flows, service-level agreement (SLA) values including priorities and minimum I/O operations per unit time;
a demand calculator configured to calculate demands for each of the external I/O flows and each of the internal I/O flows, each demand as an estimate of I/O operations per unit time demanded by a workload; and
a credit assignor configured to:
distribute credits to the internal I/O flows and the external I/O flows according to minimum I/O operations per unit time values, based on the service-level agreement values; and
distribute remaining credits to the external I/O flows and the internal I/O flows according to priority and demand of each of the external I/O flows and the internal I/O flows.
15. The storage system of claim 14 , further comprising:
one or more storage stacks or storage queues; and
the flow tracker configured to manage the internal I/O flows through the one or more storage stacks or storage queues on paths to the storage memory, based on the credits assigned to each of the internal I/O flows.
16. The storage system of claim 14 , wherein each demand is based on arrival I/O operations per unit time and total latency for serving and queuing, for each of the external I/O flows and each of the internal I/O flows.
17. The storage system of claim 14 , further comprising:
the credit assignor configured to determine credits proportional to priority, for each of the external I/O flows and each of the internal I/O flows, based on the service-level agreement values.
18. The storage system of claim 14 , further comprising:
the credit assignor configured to determine credits proportional to priority and less than calculated demand, for each of the external I/O flows and each of the internal I/O flows, wherein the priority is based on the service-level agreement values.
19. The storage system of claim 14 , further comprising:
the credit assignor configured to assign a greater number of credits to an internal I/O flow corresponding to an external I/O flow of higher priority than to an internal I/O flow corresponding to an external I/O flow of lower priority, as a result of the service-level agreement values being inherited from the external I/O flows to the internal I/O flows.
20. The storage system of claim 14 , further comprising:
the demand calculator configured to set a demand of a flow proportional to an arrival I/O operations per unit time of the flow when a total latency for serving and queuing the flow is less than or equal to a deadline for the flow based on the service-level agreement values, for each of the external I/O flows and each of the internal I/O flows; and
the demand calculator configured to set the demand of the flow proportional to a sum of the arrival I/O operations per unit time of the flow plus the total latency for serving and queuing the flow, divided by the deadline for the flow, when the total latency for serving and queuing the flow is greater than the deadline for the flow based on the service-level agreement values, for each of the external I/O flows and each of the internal I/O flows.