Performance service level agreements in multi-tenant database systems

• US 9,311,376 B2
• Filed: 05/02/2012
• Issued: 04/12/2016
• Est. Priority Date: 05/02/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Narasayya, Vivek Ravindranath, Li, Feng, Chaudhuri, Surajit
• Primary Examiner(s): Hu, Jensen
• Application Number:
  US13/461,785
• Publication Number:
  US 20130297655A1
• Time in Patent Office: 1,441 Days
• Field of Search: 707/791, 702/182, 702/186
• US Class Current: 1/1
• CPC Class Codes: G06F 11/3457 Performance evaluation by simulation
• CPC Class Codes: G06F 16/27 Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
• CPC Class Codes: H04L 41/5003 Managing SLA; Interaction between SLA and QoS
• CPC Class Codes: H04L 41/5009 Determining service level performance parameters or violations of service level contracts, e.g. violations of agreed response time or mean time between failures [MTBF]

Abstract

Various technologies described herein pertain to evaluating service provider compliance with terms of a performance service level agreement (SLA) for a tenant in a multi-tenant database system. The terms of the performance SLA can set a performance criterion as though a level of a resource of hardware of the multi-tenant database system is dedicated to the tenant. An actual performance metric of the resource can be tracked for a workload of the tenant. Further, a baseline performance metric of the resource can be determined for the workload of the tenant. The baseline performance metric can be based on a simulation as though the level of the resource as set in the performance SLA is dedicated to the workload of the tenant. Moreover, the actual performance metric can be compared with the baseline performance metric to evaluate compliance with the performance SLA.

Claims

1. A method of managing a multi-tenant database system, comprising:

tracking an actual performance metric of a resource of hardware of the multi-tenant database system for a workload of a tenant, the resource of the hardware of the multi-tenant database system being shared with at least one disparate workload of at least one disparate tenant, the multi-tenant database system having a performance service level agreement (SLA) for the tenant, terms of the performance SLA set a performance criterion as though a level of the resource of the hardware of the multi-tenant database system is dedicated for the tenant;

simulating utilization of a dedicated resource of hardware, the utilization of the dedicated resource of the hardware simulated based on;

the workload of the tenant; and

the level of the resource of the hardware of the multi-tenant database system set in the performance SLA as being dedicated to the workload of the tenant;

determining a baseline performance metric of the resource of the hardware of the multi-tenant database system for the workload of the tenant based on the utilization of the dedicated resource of the hardware as simulated;

comparing the actual performance metric with the baseline performance metric to evaluate a performance degradation ( ); and

allocating the resource of the hardware of the multi-tenant database system to the workload of the tenant based on the performance degradation.

2. The method of claim 1, the resource of the hardware of the multi-tenant database system further being allocated to the workload of the tenant based on a penalty function that penalizes the performance degradation, wherein the penalty function is specified by the performance SLA.

3. The method of claim 1, further comprising:

metering compliance with the performance SLA independently during different time intervals, wherein a predefined duration is set for the different time intervals by the performance SLA; and

selectively compensating the tenant for each of the different time intervals based on the independently metered compliance with the performance SLA.

4. The method of claim 1, wherein the level of the resource of the hardware of the multi-tenant database system specified in the performance SLA is at least one of a number of IOs per second or a latency for IOs.

5. The method of claim 1, wherein the level of the resource of the hardware of the multi-tenant database system specified in the performance SLA is an amount of CPU utilization.

6. The method of claim 1, wherein the level of the resource of the hardware of the multi-tenant database system specified in the performance SLA is an amount of working memory.

7. The method of claim 1, wherein the level of the resource of the hardware of the multi-tenant database system specified in the performance SLA is an amount of buffer pool memory.

8. The method of claim 7, wherein the actual performance metric of the buffer pool memory is an actual hit rate of the buffer pool memory, and the baseline performance metric of the buffer pool memory is a baseline hit rate of the buffer pool memory, and the performance degradation is a hit rate degradation;

and

allocating the resource of the hardware of the multi-tenant database system to the workload of the tenant based on the performance degradation further comprising;

evicting a selected page from the buffer pool memory of the multi-tenant database system based on the hit rate degradation.

9. The method of claim 8, further comprising:

measuring a number of pages accessed by the workload of the tenant;

for each of the pages accessed by the workload of the tenant, determining whether the page is found in the buffer pool memory; and

computing the actual hit rate as a fraction of the pages accessed by the workload of the tenant that are found in the buffer pool memory.

10. The method of claim 8, further comprising:

simulating page replacement that would have been carried out for the workload of the tenant given the amount of the buffer pool memory being dedicated to the tenant, wherein the page replacement is simulated in a baseline buffer pool based on a page replacement policy;

measuring a number of pages accessed by the workload of the tenant;

for each of the pages accessed by the workload of the tenant, determining whether the page is found in the baseline buffer pool; and

computing the baseline hit rate as a fraction of the pages accessed by the workload of the tenant that are found in the baseline buffer pool.

11. The method of claim 10, wherein the baseline buffer pool comprises page IDs of the pages accessed by the workload of the tenant and lacks contents of the pages accessed by the workload of the tenant.

12. The method of claim 1, wherein the performance SLA for the tenant in the multi-tenant database system specifies an amount of buffer pool memory, a performance metric type, a predefined duration of time for a time interval at which compliance with the performance SLA is metered, pricing information, and a penalty function.

13. The method of claim 1, further comprising allocating the resource of the hardware of the multi-tenant database system between the workload of the tenant and the at least one disparate workload of the at least one disparate tenant based on:

the performance degradation of the tenant;

a penalty function specified in the performance SLA for the tenant;

at least one performance degradation of the at least one disparate tenant; and

at least one penalty function specified in at least one disparate performance SLA for the at least one disparate tenant.

14. A multi-tenant database system, comprising:

a resource of hardware;

at least one processor; and

memory that comprises computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including;

evaluating, for a workload of a tenant and disparate workloads of disparate tenants, performance degradations between actual performance metrics of the resource of the hardware and baseline performance metrics of the resource of the hardware; and

allocating the resource of the hardware between the workload of the tenant and the disparate workloads of the disparate tenants based on the performance degradations for the workload of the tenant and the disparate workloads of the disparate tenants;

wherein a performance service level agreement (SLA) for the tenant and disparate performance SLAs for the disparate tenants comprise terms that set performance criteria as though corresponding levels of the resource of the hardware are respectively dedicated for the tenant and each of the disparate tenants; and

wherein the baseline performance metrics of the resource of the hardware for the workload of the tenant and the disparate workloads of the disparate tenants are based on simulations of utilizations of the levels of the resource of the hardware respectively dedicated for the tenant and each of the disparate tenants set in the performance SLA and the disparate performance SLAs for the workload of the tenant and the disparate workloads of the disparate tenants.

15. The multi-tenant database system of claim 14, wherein the resource of the hardware is buffer pool memory, and the memory further comprising computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including:

selecting a page eviction from the buffer pool memory based on the performance degradation for the workload of the tenant and the disparate workloads of the disparate tenants.

16. The multi-tenant database system of claim 15, the memory further comprising computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including:

selecting a page for eviction from the buffer pool memory further based on a penalty function specified in the performance SLA for the tenant and disparate penalty functions specified in the disparate performance SLAs for the disparate tenants.

17. The multi-tenant database system of claim 16, the memory further comprising computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including:

determining an expected penalty incurred if the page is evicted from the buffer pool memory based upon an estimation of a hit rate degradation, wherein the page is selected for eviction based on the expected penalty.

18. The multi-tenant database system of claim 16, the memory further comprising computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including:

selecting the page for eviction from the buffer pool memory based on a likelihood of the page being accessed by one of;

the workload of the tenant or one of the workloads of the disparate tenants.

19. The multi-tenant database system of claim 14, the memory further comprising computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to perform acts including:

evaluating service provider compliance with the terms of the performance SLA for the tenant and the disparate performance SLAs for the disparate tenants.

20. A computer-readable storage medium including computer-executable instructions that, when executed by a processor, cause the processor to perform acts including:

measuring a number of pages accessed by a workload of a tenant in a multi-tenant database system;

for each of the pages accessed by the workload of the tenant, determining whether the page is found in buffer pool memory of the multi-tenant database system, the buffer pool memory of the multi-tenant database system being shared with a concurrently executing workload of a disparate tenant;

computing an actual hit rate as a fraction of the pages accessed by the workload of the tenant that are found in the buffer pool memory;

simulating page replacement that would have been carried out for the workload of the tenant in dedicated buffer pool memory, the page replacement simulated based on the workload of the tenant and an amount of the dedicated buffer pool memory specified in a performance service level agreement (SLA) as being dedicated to the tenant, wherein the page replacement is simulated in a baseline buffer pool based on a page replacement policy;

for each of the pages accessed by the workload of the tenant, determining whether the page is found in the baseline buffer pool;

computing a baseline hit rate as a fraction of the pages accessed by the workload of the tenant that are found in the baseline buffer pool;

comparing the actual hit rate with the baseline hit rate to evaluate a hit rate degradation; and

evicting a selected page from the buffer pool memory of the multi-tenant database system based on the hit rate degradation and a penalty function specified in the performance SLA.