Method of distributing load amongst two or more computer system resources

1. A method comprising:

determining, for each of a plurality of system resources, a modeled response time for a given load, wherein the modeled response time is calculated using an application service demand multiplied by a sum of 1 and an average queue length, wherein the application service demand comprises the demand of a particular application executing on each of the plurality of system resources and the average queue length is a queue length averaged over all the plurality of system resources and applications;

determining, for each of the system resources, a Service Level Agreement (SLA) miss value by subtracting the modeled response time of each of the system resources for a given load from a predetermined SLA of each of the system resources; and

distributing said load to the system resource having the greatest SLA miss value.

2. A method according to claim 1 wherein determining the modeled response time further comprises determining the modeled response time by analytical modeling using a queuing network model.

3. A method according to claim 1 wherein each system resource is a server including a central processing unit and one or more storage devices.

4. A method according to claim 1 including receiving a request associated with the given load.

5. A method according to claim 4 wherein the request is a web server request.

6. A method according to claim 1 wherein the load is distributed in accordance with a single Service Level Agreement (SLA) which is the same for all system resources.

7. A method according to claim 1 wherein distributing further comprises distributing the load in accordance with a first Service Level Agreement (SLA) associated with a first one of the system resources and a second Service Level Agreement (SLA) associated with a second one of the system resources.

8. A method according to claim 1 wherein the load is distributed to the system resource by stopping a process on one of the system resources and migrating the process to a different system resource.

9. A method according to claim 1 further including monitoring each system resource over a measurement interval to generate a measured response time, and calibrating each modeled response time in accordance with a respective measured response time.

10. A method according to claim 1 further including monitoring each system resource to generate one or more application metrics, and modeling the response time in accordance with the one or more application metrics.

11. A method according to claim 10 wherein the one or more application metrics include a parameter indicating the amount of time during a measurement interval in which a process is alive.

12. A method according to claim 10 wherein the one or more application metrics include a parameter indicating the amount of time during a measurement interval in which a process is consuming resources.

13. A method according to claim 10 wherein the one or more application metrics include a parameter indicating the number of processes that completed during a measurement interval.

14. A computer system including:

a processor configured to execute a load distributor;

two or more system resources; and

two or more agents, each agent configured to model a response time of the system resources for a given load, wherein the modeled response time is calculated using an application service demand multiplied by a sum of 1 and an average queue length, wherein the application service demand comprises the demand of a particular application executing on each of the plurality of system resources and the average queue length is a queue length averaged over all the plurality of system resources and applications;

wherein the load distributor is configured to:

receive the modeled response time of each of the system resources from the agents;

determine, for each of the system resources, a Service Level Agreement (SLA) miss value by subtracting the response time of each of the system resources for a given load from a predetermined SLA of each of the system resources; and

distribute said load to the system resource having the greatest SLA miss value.

15. A non-transitory computer-readable storage medium comprising software that, when executed by a processor, causes the processor to:

determine, for each of a plurality of resources, a modeled response time for a given load, wherein the modeled response time is calculated using an application service demand multiplied by a sum of 1 and an average queue length, wherein the application service demand comprises the demand of a particular application executing on each of the plurality of system resources and the average queue length is a queue length averaged over all the plurality of system resources and applications;

determine, for each of the resources, a Service Level Agreement (SLA) miss value by subtracting the modeled response time of each of the resources for a given load from a predetermines SLA of each of the resources; and

distribute said load to the one of the resources having the greatest SLA miss value.

16. A method comprising:

determining, for each of a plurality of system resources, a modeled response time for a given load, wherein the modeled response time is calculated using an application service demand multiplied by a sum of 1 and an average queue length, wherein the application service demand comprises the demand of a particular application executing on each of the plurality of system resources and the average queue length is a queue length averaged over, all the plurality of system resources and applications;

distributing the load to the system resource with the lowest modeled response time.