SATURATION DETECTION AND ADMISSION CONTROL FOR STORAGE DEVICES

• US 20120054329A1
• Filed: 08/27/2010
• Published: 03/01/2012
• Est. Priority Date: 08/27/2010
• Status: Active Grant

• Current Assignee: Vmware LLC (Broadcom, Inc.)
• Original Assignee: VMware, Inc. (Broadcom, Inc.)
• Inventors: GULATI, Ajay, SHANMUGANATHAN, Ganesha, AHMAD, Irfan
• Granted Patent: US 9,467,505 B2
• Time in Patent Office: Days
• US Class Current: 709/224
• CPC Class Codes: H04L 43/0852 Delays
• CPC Class Codes: H04L 43/16 Threshold monitoring
• CPC Class Codes: H04L 43/50 Testing arrangements
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

Maximum throughput of a storage unit, and workload and latency values of the storage unit corresponding to a predefined fraction of the maximum throughput are estimated based on workloads and latencies that are monitored on the storage unit. The computed metrics are usable in a variety of different applications including admission control, storage load balancing, and enforcing quality of service in a shared storage environment.

Claims

1. A method of estimating throughput of a storage unit, comprising:

monitoring a workload on the storage unit and a latency of the storage unit at multiple points in time over a period of time; and

determining a maximum throughput of the storage unit based on a linear relationship between the monitored workloads and the monitored latencies.

2. The method of claim 1, wherein said determining comprises:

performing a linear fit between the monitored workloads and the monitored latencies,

wherein the maximum throughput of the storage unit is determined as an inverse of the slope of the linear fit.

3. The method of claim 2, wherein the linear fit is performed between monitored workloads that are greater than a predetermined workload and the monitored latencies that correspond to such monitored workloads.

4. The method of claim 1, wherein the workload on the storage unit is monitored by monitoring outstanding IOs to the storage unit.

5. The method of claim 1, further comprising:

estimating a latency of the storage unit operating at a predetermined fraction of the maximum throughput.

11. The method of claim 1, further comprising:

detecting an idle period for the storage unit; and

injecting a controlled workload into the storage unit over the period of time.

12. The method of claim 11, wherein the control workload includes IO requests that are generated repeatedly over multiple time intervals and the number of IO requests generated at each subsequent time interval increases.

6. A method of controlling admissions of a workload into a storage unit, comprising:

estimating a new throughput that would result if the workload is admitted;

computing a threshold latency corresponding to a predefined fraction for the new throughput relative to a maximum throughput of the storage unit;

estimating a total latency that would result if the workload is admitted;

comparing the estimated total latency with the threshold latency; and

admitting the workload if the estimated total latency is less than the threshold latency.

7. The method of claim 6, wherein the predefined fraction is 100%.

8. The method of claim 6, wherein the predefined fraction is less than 100%.

9. The method of claim 6, wherein the maximum throughput of the storage unit is determined as an inverse of a slope of a line that characterizes a relationship between workload on the storage unit and latency of the storage unit.

10. The method of claim 6, further comprising:

rejecting the workload if the estimated total latency is greater than the threshold throughput.

13. A method of load balancing workloads across storage units, comprising:

selecting a workload for migration to a destination storage unit;

determining whether or not migration of the selected workload to the destination storage unit will cause the destination storage unit to reach a predefined fraction of a saturation workload; and

migrating the selected workload to the destination storage unit if the predefined fraction of the saturation workload of the storage unit will not be reached.

14. The method of claim 13, wherein the predefined fraction is 100%.

15. The method of claim 13, wherein the predefined fraction is less than 100%.

16. The method of claim 13, wherein the saturation workload of the storage unit is determined based in part on an inverse of a slope of a line that characterizes a relationship between monitored workloads on the storage unit and monitored latencies of the storage unit.

17. The method of claim 13, wherein the storage unit operates at a maximum throughput at workloads greater than or equal to the saturation workload.

18. In a system having a plurality of hosts sharing a common storage unit, a method carried out by each of the hosts to enforce a quality of service policy, comprising:

determining an average latency across all of the hosts;

comparing the average latency with a threshold latency; and

adjusting IO issue queue size of the host,

wherein the threshold latency is determined as latency of the common storage unit operating at a predefined fraction of maximum throughput.

19. The method of claim 18, wherein the predefined fraction is 100%.

20. The method of claim 18, wherein the predefined fraction is less than 100%.

21. The method of claim 18, wherein the maximum throughput of the common storage unit is determined based on an inverse of a slope of a line that characterizes a relationship between monitored workloads on the common storage unit and monitored latencies of the common storage unit.

22. The method of claim 18, wherein the IO issue queue size of the host is adjusted based in part on assigned shares of the host.