Method for dividing computations

• US 20070260669A1
• Filed: 07/10/2007
• Published: 11/08/2007
• Est. Priority Date: 06/20/2002
• Status: Active Grant

• Current Assignee: JP Morgan Chase Bank, N.A. (JP Morgan Chase & Co)
• Original Assignee: JP Morgan Chase & Co
• Inventors: Sulzhyk, Roman, Neiman, Steven
• Granted Patent: US 7,415,496 B2
• Time in Patent Office: Days
• US Class Current: 709/201
• CPC Class Codes: G06F 2209/5017 Task decomposition
• CPC Class Codes: G06F 9/5038 considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration scheduling strategies G06F9/4881 and subgroups
• CPC Class Codes: G06F 9/505 considering the load

Abstract

In certain aspects, the invention features methods that include receiving a parent job configured to produce one or more descendant jobs, and scheduling computation of the parent job on a node computing device that is one of a plurality of node computing devices of a distributed computing system. Such an aspect also includes selectively rescheduling computation of a job other than a parent job from any one of the node computing devices to another of the node computing devices, and preventing rescheduling of the parent job unless each of the descendant jobs is completed or terminated.

Claims

1. A method, comprising:

receiving, for computation by a node computing device of a distributed computing system, a parent job configured to produce one or more descendant jobs, wherein said node computing device is one of a plurality of node computing devices of said distributed computing system;

scheduling computation of said parent job on said node computing device;

selectively rescheduling computation of a job other than said parent job from any one of said plurality of node computing devices to another of said node computing devices; and

preventing rescheduling of said parent job unless each of said descendant jobs is completed or terminated.

2. The method of claim 1, said distributed computing system further comprising a persistent data storage queue in communication with said node computing device, wherein a minimum availability of said distributed computing system is defined by an availability of said persistent data storage;

and

wherein said method further comprises;

storing a descendant output from each of said descendant jobs in said persistent queue for retrieval by said node computing device processing said parent job; and

accessing said persistent queue to retrieve said descendant output for use in computation of said parent job.

3. The method of claim 1, wherein none of said node computing devices is available for computation at a time when said parent job is scheduled for computation.

4. The method of claim 1, further comprising sending for computation each descendant job to a node computing device other than said node computing device processing said parent job.

5. The method of claim 1, wherein said parent job comprises meta-information comprising an instruction to divide one or more of said descendant jobs from said parent job for scheduling by a scheduler server and processing by at least one of said node computing devices.

6. The method of claim 1, further comprising terminating each of said descendant jobs upon termination of said parent job.

7. The method of claim 1, wherein each of said node computing devices provides to a scheduler server an availability status.

8. The method of claim 1, further comprising receiving said parent job from an application running on a local computing device.

9. The method of claim 8, further comprising providing an output of said parent job for retrieval by said application.

10. The method of claim 1, further comprising storing a descendant output from at least one of said descendant jobs in a cache for use by another of said descendant jobs or said parent job.

11. The method of claim 1, wherein said descendant job comprises meta-information comprising an identification of a compute function to be used to perform a computation for said descendant job.