Method for improving the performance of high performance computing applications on Cloud using integrated load balancing

1. A computer-implemented method of load balancing processing resources, said method comprising:

calculating, by a first computing device located in an application executing environment, an expected cost set associated with an application-specific task of an application executing on a processing resource in a cloud computing environment, said first computing device being outside said cloud computing environment, said expected cost set comprising an expected computational cost and an expected communication cost;

communicating, by said first computing device, said expected cost set from said first computing device to a second computing device located in a cloud management system;

retrieving, by said second computing device, resource mapping of applications currently executing in said cloud computing environment;

assigning, by said second computing device, said application-specific task to a specific computational resource in said cloud computing environment based on said expected cost set and said resource mapping of applications currently executing in said cloud computing environment;

determining, by said second computing device, a task to VM (virtual machine) assignment based on said assigning said application-specific task to said specific computational resource in said cloud computing environment;

transferring, by said second computing device, information concerning said task to VM assignment from said cloud management system to said application executing on said processing resource in said cloud computing environment;

migrating, by said second computing device, VMs of existing applications onto different processing resources in said cloud computing environment based on said assigning said application-specific task to said specific computational resource in said cloud computing environment; and

redistributing, by said first computing device, data and said application-specific task of said application based on said transferring said information concerning said task to VM assignment from said cloud management system to said application executing on said processing resource of said cloud computing environment.

2. The computer-implemented method of load balancing according to claim 1 , said expected cost set associated with said application-specific task being represented in a directed task graph form.

3. The computer-implemented method of load balancing according to claim 2 , said directed task graph form comprising:

vertices that correspond to a task;

edges that correspond to a task dependency;

vertex weights that correspond to said expected computational cost; and

edge weights that correspond to said expected communication cost.

4. The computer-implemented method of load balancing according to claim 2 , said directed task graph form being generated by an application-specific coarsening of an original problem domain.

5. The computer-implemented method of load balancing according to claim 1 , said method of load balancing being performed a plurality of times in an iterative fashion.

6. A computer-implemented method of load balancing processing resources, said method comprising:

calculating, by a first computing device located in an application executing environment, an expected computational cost and an expected communication cost associated with an application-specific task of an application executing on a processing resource in a cloud computing environment, said first computing device being outside said cloud computing environment;

communicating, by said first computing device, said expected computational cost and said expected communication cost from said processing resource to a second computing device located in a cloud management system;

retrieving, by said second computing device, resource mapping of applications currently executing in said cloud computing environment;

assigning, by said second computing device, said application-specific task to a specific computational resource in said cloud computing environment based on said expected computational cost, said expected communication cost, and said resource mapping of said applications currently executing in said cloud computing environment;

determining, by said second computing device, a task to VM (virtual machine) assignment based on said assigning of said application-specific task to said specific computational resource;

transferring, by said second computing device, said task to VM assignment from said cloud management system to said application executing on said processing resource in said cloud computing environment;

migrating, by said second computing device, VMs of existing applications onto different processing resources in said cloud computing environment based on said assigning said application-specific task to said specific computational resource; and

redistributing, by said first computing device, data and said application-specific task of said application based on said transferring said task to VM assignment from said cloud management system to said application executing on said processing resource of said cloud computing environment.

7. The computer-implemented method of load balancing according to claim 6 , said expected computational cost and said expected communication cost associated with said application-specific task being represented in a directed task graph form.

8. The computer-implemented method of load balancing according to claim 7 , said directed task graph form comprising:

vertices that correspond to a task;

edges that correspond to a task dependency;

vertex weights that correspond to said expected computational cost; and

edge weights that correspond to said expected communication cost.

9. The computer-implemented method of load balancing according to claim 7 , said directed task graph form being generated by an application-specific coarsening of an original problem domain.

10. The computer-implemented method of load balancing according to claim 6 , said method of load balancing being performed a plurality of times in an iterative fashion.

11. The computer-implemented method of load balancing according to claim 6 , said assigning said application-specific task to said specific computational resource in said cloud computing environment being further based on modeling resource capacity from monitored historical values.

12. The computer-implemented method of load balancing according to claim 6 , said assigning said application-specific task to said specific computational resource in said cloud computing environment being further based on task information from multiple applications.

13. A computer-implemented method of load balancing processing resources, said method comprising:

calculating, by a first computing device located in an application executing environment, an expected computational cost and an expected communication cost associated with an application-specific task of an application executing on a processing resource in a cloud computing environment, said first computing device being outside said cloud computing environment;

communicating, by said first computing device, said expected computational cost and said expected communication cost from said processing resource to a second computing device located in a cloud management system;

retrieving, by said second computing device, resource mapping of applications currently executing in said cloud computing environment;

modeling resource capacity from monitored historical values and task information from multiple applications;

assigning, by said second computing device, said application-specific task to a specific computational resource in said cloud computing environment based on said expected computational cost, said expected communication cost, and resource availability for applications currently executing in said cloud computing environment and based on said modeling said resource capacity from monitored historical values and task information from multiple applications;

determining, by said second computing device, a task to VM (virtual machine) assignment based on said assigning of said application-specific task to said specific computational resource;

transferring, by said second computing device, said task to VM assignment from said cloud management system to said application executing on said processing resource in said cloud computing environment;

migrating, by said second computing device, VMs of existing applications onto different processing resources in said cloud computing environment based on said assigning said application-specific task to said specific computational resource; and

redistributing, by said first computing device, data and said application-specific task of said application based on said transferring said task to VM assignment from said cloud management system to said application executing on said processing resource of said cloud computing environment.

14. The computer-implemented method of load balancing according to claim 13 , said expected computational cost and said expected communication cost associated with said application-specific task being represented in a directed task graph form.

15. The computer-implemented method of load balancing according to claim 14 , said directed task graph form comprising:

vertices that correspond to a task;

edges that correspond to a task dependency;

vertex weights that correspond to said expected computational cost; and

edge weights that correspond to said expected communication cost.

16. The computer-implemented method of load balancing according to claim 14 , said directed task graph form being generated by an application-specific coarsening of an original problem domain.

17. The computer-implemented method of load balancing according to claim 13 , said method of load balancing being performed a plurality of times in an iterative fashion.