VIRTUAL MACHINE DEMAND ESTIMATION

• US 20110072138A1
• Filed: 09/21/2009
• Published: 03/24/2011
• Est. Priority Date: 09/21/2009
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Whalley, Ian N., Steinder, Malgorzata, Canturk, Isci, Hanson, James E., Kephart, Jeffrey O.
• Granted Patent: US 9,037,717 B2
• Time in Patent Office: Days
• US Class Current: 709/226
• CPC Class Codes: G06F 2209/504 Resource capping
• CPC Class Codes: G06F 9/5011 the resources being hardware resources other than CPUs, Servers and Terminals
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A method for use in a system in which computational entities are distributed across physical computing resources to place the entities on the resources includes estimating actual resource demand for each entity on each resource based on application resource usage data collected from a data source external from the entity, computing a best allocation of the resources to the entities from the estimated actual resource demand for each entity and distributing the resources to the entities in accordance with the computed best allocation.

Claims

1. A method for use in a system in which computational entities are distributed across physical computing resources, the method comprising:

establishing a utilization target for each resource;

estimating actual resource demand for each entity on each resource based on application resource usage data collected from a data source external from the entity;

computing a best allocation of the resources to the entities from the estimated actual resource demand for each entity such that, upon distribution of the resources to the entities, the resources will substantially satisfy the estimated actual resource demand for each entity without exceeding the established utilization target; and

distributing the resources to the entities in accordance with the computed best allocation in a single distribution iteration.

2. The method according to claim 1, wherein the estimating comprises:

obtaining a map of the entities relative to the resources;

obtaining resource usage metrics for each entity and resource;

obtaining resource capacities; and

computing the estimated actual resource demand based on the map of the entities relative to the resources, the resource usage metrics and the resource capacities.

3. The method according to claim 2, wherein the resource capacities are time-varying.

4. The method according to claim 2, wherein the computing of the estimated actual resource demand comprises:

computing estimated actual resource demand specific to CPU resources;

computing estimated actual resource demand specific to memory resources;

computing estimated actual resource demand specific to storage I/O resources; and

computing estimated actual resource demand specific to network I/O resources.

5. The method according to claim 4, wherein the computing of the estimated actual resource demand specific to CPU resources comprises:

obtaining CPU resource usage metrics for each entity and resource;

obtaining CPU resource capacities for each resource; and

computing actual resource demand specific to CPU resources for each entity from the CPU resource usage metrics and the CPU resource capacities.

6. The method according to claim 5, wherein the CPU resource usage metrics comprise:

CPU cycles during which the entity was actually granted the use of the CPU resources;

CPU cycles during which the entity was ready to use the CPU resources;

CPU cycles during which the entity was waiting on a resource other than the CPU; and

CPU cycles during which the entity was idle.

7. The method according to claim 6, wherein the CPU resource usage metrics further comprise CPU cycles during which the resource performed system-related operations.

8. The method according to claim 5, wherein the CPU resource capacities comprise available CPU cycles aggregated over all available logical processing units enclosed by the resource.

9. The method according to claim 4, wherein the computing of the estimated actual resource demand specific to memory resources comprises:

obtaining memory resource usage metrics for each entity and resource;

obtaining memory resource capacities for each entity and resource; and

computing actual resource demand specific to memory resources for each entity from the memory resource usage metrics and the memory resource capacities.

10. The method according to claim 9, wherein the memory resource usage metrics comprise:

an amount of memory that is active for each entity;

an amount of memory that is mapped for each entity;

an amount of memory that is swapped for each entity;

an amount of memory that is shared with other entities; and

an amount of reserved memory.

11. The method according to claim 9, wherein the memory resource capacities comprise available and configured memory for each resource.

12. The method according to claim 4, wherein the computing of the estimated actual resource demand specific to storage I/O resources comprises:

obtaining storage resource usage metrics for each entity and resource;

obtaining storage resource capacities for each entity and resource; and

computing actual resource demand specific to storage I/O resources for each entity from the storage resource usage metrics and the storage resource capacities.

13. The method according to claim 12, wherein the storage resource usage metrics comprise:

storage read and write rates for each entity;

storage read and write requests for each entity;

storage read and write latencies for each entity and resource; and

storage device queue read and write latencies for each resource.

14. The method according to claim 12, wherein the storage resource capacities comprise storage bandwidth for each resource.

15. The method according to claim 4, wherein the computing of the estimated actual resource demand specific to network I/O resources comprises:

obtaining network resource usage metrics for each entity and resource;

obtaining network resource capacities for each entity and resource; and

computing actual resource demand specific to network I/O resources for each entity from the network resource usage metrics and the network resource capacities.

16. The method according to claim 15, wherein the network resource usage metrics comprise:

network packet transfer and receive rates for each entity and resource;

packet transfer and receive latencies for each entity; and

transmit and receive queue lengths and latencies for each resource.

17. The method according to claim 15, wherein the network resource capacities comprise:

available network bandwidth for each resource; and

configured network bandwidth for each entity.

18. The method according to claim 2, further comprising forecasting future demand based on the estimated actual resource demand.

19. A method for estimating actual resource demand for running entities, the method comprising:

obtaining a map of the entities relative to resources;

obtaining resource usage metrics for each entity and resource, the resource usage metrics for each resource including established utilization target;

obtaining resource capacities; and

based on the map of the entities relative to the resources, the resource usage metrics and the resource capacities;

estimating actual resource demand for each entity on each resource,

computing a best allocation of the resources to the entities from the estimated actual resource demand for each entity such that, upon distribution of the resources to the entities, the resources will substantially satisfy the estimated actual resource demand for each entity without exceeding the established utilization target, and

distributing the resources to the entities in accordance with the computed best allocation in a single distribution iteration.

20. The method according to claim 19, wherein the estimating actual resource demand comprises:

computing estimated actual resource demand specific to CPU resources;

computing estimated actual resource demand specific to memory resources;

computing estimated actual resource demand specific to storage I/O resources; and

computing estimated actual resource demand specific to network I/O resources.

21. The method according to claim 19, further comprising forecasting future demand based on the estimated actual resource demand.

22. A system, in which computational entities are distributed across physical computing resources to place the entities on the resources, the system including a processor and a computer readable medium coupled to the processor having executable instructions stored thereon, which, when executed instruct the processor to establish a utilization target for each resource, to estimate actual resource demand for each entity on each resource based on application resource usage data collected from a data source external from the entity, to compute a best allocation of the resources to the entities from the estimated actual resource demand for each entity, such that, upon distribution of the resources to the entities, the resources will substantially satisfy the estimated actual resource demand for each entity without exceeding the established utilization target, and to distribute the resources to the entities in accordance with the computed best allocation in a single distribution iteration.