Stream processing with dynamic event routing

• US 10,055,691 B2
• Filed: 05/12/2015
• Issued: 08/21/2018
• Est. Priority Date: 09/08/2014
• Status: Active Grant

• Current Assignee: Pivotal Software, Inc. (Broadcom, Inc.)
• Original Assignee: Pivotal Software, Inc. (Broadcom, Inc.)
• Inventors: Brand, Michael, Adams, Lyndon John, Brown, David Russell, Ng, Kee Siong
• Primary Examiner(s): Misir, Dave
• Application Number:
  US14/710,352
• Publication Number:
  US 20160071011A1
• Time in Patent Office: 1,197 Days
• Field of Search: 706 12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: H04L 41/16 using machine learning or artificial intelligence
• CPC Class Codes: H04L 45/08 Learning-based routing, e.g. using neural networks or artificial intelligence

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for routing events of an event stream. One of the methods includes operations of receiving, by a router, events of an event stream; providing each event, by the router, to a respective local modeler selected by the router according to an initial routing strategy, the respective local modeler being selected from multiple local modelers; aggregating, by each local modeler in parallel, information associated with each event received by the local modeler to generate aggregated information; providing, to a central modeler, the aggregated information generated by the one or more local modelers; determining, by the central modeler, parameters of a machine learning model using the aggregated information received by the central modeler and generating an updated routing strategy based on the parameters of the machine learning model; and providing the updated routing strategy to the router.

Claims

1. A method comprising:

receiving, by a first router executing on a system of one or more computers, events of an event stream;

providing each event in the event stream received by the first router to a respective first local modeler selected by the first router according to an initial routing strategy, the respective first local modeler being selected from a plurality of first local modelers executing on the system and processing events in the event stream in parallel;

allowing fast-data computation with computation-intensive processing by;

generating, by each first local modeler of the plurality of first local modelers at least partially in parallel, respective partial aggregated information computed from values of respective events received by the first local model from the first router;

providing, by each first local modeler to a central modeler executing on the system, the respective partial aggregated information generated by the first local modeler of the plurality of first local modelers;

computing, by the central modeler, aggregated information using instances of partial aggregated information computed at least partially in parallel by the plurality of first local modelers; and

determining, by the central modeler, a plurality of parameters of a machine learning model using the aggregated information computed from the instances of partial aggregated information generated by the plurality of first local modelers, wherein the parameters of the machine learning model determine an assignment of events to respective sub-populations of events;

providing, by the central modeler, the plurality of parameters of the machine learning model to each of the first local modelers;

generating, by the central modeler, an updated routing strategy that assigns each event belonging to a particular sub-population of events according to the parameters of the machine learning model to a particular second local modeler of a plurality of second local modelers;

providing, by the central modeler while the first local modelers continue to process events in the event stream, the updated routing strategy to a second router that routes events processed by the plurality of first local modelers;

processing, by the first local modelers, events in the event stream according to the parameters of the machine learning model to assign each event to a respective sub-population of events;

providing, by the first local modelers, the processed events to the second router; and

providing, by the second router, each processed event received from the first local modelers to a respective second local modeler according to a sub-population of events assigned by the first local modeler based on the updated routing strategy computed by the central modeler.

2. The method of claim 1, wherein determining, by the central modeler, the plurality of parameters of the machine learning model using the aggregated information comprises performing a clustering process using aggregated information provided by the plurality of first local modelers.

3. The method of claim 2, wherein the clustering process is a k-means clustering process or a mixture model process that identifies clusters of events or mixtures of events each centered around one of a multiple of means.

4. The method of claim 1, wherein the updated routing strategy specifies that the second router should route the events by one or more data elements in a data tuple defining each event.

5. The method of claim 1, further comprising:

determining, by the central modeler, from the updated routing strategy, respective context data that should be stored in operational memory of each of the plurality of second local modelers; and

obtaining, by each second local modeler of the plurality of second local modelers, the respective context data for storage in operational memory of the second local modeler.

6. The method of claim 5, further comprising:

obtaining, by each of the second local modelers, the respective context data from one or more other local modelers or from a storage system that has all of the context data.

7. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

receiving, by a first router executing on the one or more computers, events of an event stream;

providing each event in the event stream received by the first router to a respective first local modeler selected by the first router according to an initial routing strategy, the respective first local modeler being selected from a plurality of first local modelers executing on the one or more computers and processing events in the event stream in parallel;

allowing fast-data computation with computation-intensive processing by;

generating, by each first local modeler of the plurality of first local modelers at least partially in parallel, respective partial aggregated information computed from values of respective events received by the first local modeler from the first router;

providing, by each first local modeler to a central modeler executing on the one or more computers, the respective partial aggregated information generated by the first local modeler of the plurality of first local modelers, the central modeler executing on the one or more computers;

computing, by the central modeler, aggregated information using instances of partial aggregated information computed at least partially in parallel by the plurality of first local modelers; and

determining, by the central modeler, a plurality of parameters of a machine learning model using the aggregated information computed from the instances of partial aggregated information generated by the plurality of first local modelers, wherein the parameters of the machine learning model determine an assignment of events to respective sub-populations of events;

providing, by the central modeler, the plurality of parameters of the machine learning model to each of the first local modelers;

generating, by the central modeler, an updated routing strategy that assigns each event belonging to a particular sub-population of events according to the parameters of the machine learning model to a particular second local modeler of a plurality of second local modelers executing on the one or more computers;

providing, by the central modeler while the first local modelers continue to process events in the event stream, the updated routing strategy to a second router that routes events processed by the plurality of first local modelers, wherein the second router executes on the one or more computers;

processing, by the first local modelers, events in the event stream according to the parameters of the machine learning model to assign each event to a respective sub-population of events;

providing, by the first local modelers, the processed events to the second router; and

providing, by the second router, each processed event received from the first local modelers to a respective second local modeler according to a sub-population of events assigned by the first local modeler based on the updated routing strategy computed by the central modeler.

8. The system of claim 7, wherein determining, by the central modeler, the plurality of parameters of the machine learning model using the aggregated information comprises performing a clustering process using aggregated information provided by the plurality of first local modelers.

9. The system of claim 8, wherein the clustering process is a k-means clustering process or a mixture model process that identifies clusters of events or mixtures of events each centered around one of a multiple of means.

10. The system of claim 7, wherein the updated routing strategy specifies that the second router should route the events by one or more data elements in a data tuple defining each event.

11. The system of claim 7, wherein the operations further comprise:

determining, by the central modeler, from the updated routing strategy, respective context data that should be stored in operational memory of each of the plurality of second local modelers; and

obtaining, by each second local modeler of the plurality of second local modelers, the respective context data for storage in operational memory of the second local modeler.

12. The system of claim 11, wherein the operations further comprise:

obtaining, by each of the second local modelers, the respective context data from one or more other local modelers or from a storage system that has all of the context data.

13. One or more non-transitory computer-readable storage media encoded with instructions that, when executed by a stream processing system comprising one or more computers, cause the one or more computers to perform operations comprising:

receiving, by a first router executing on the stream processing system, events of an event stream;

providing each event in the event stream received by the first router to a respective first local modeler selected by the first router according to an initial routing strategy, the respective first local modeler being selected from a plurality of first local modelers executing on the stream processing system and processing events in the event stream in parallel;

allowing fast-data computation with computation-intensive processing by;

generating, by each first local modeler of the plurality of first local modelers at least partially in parallel, respective partial aggregated information computed from values of respective events received by the first local modeler from the first router;

providing, by each first local modeler to a central modeler executing on the stream processing system, the respective partial aggregated information generated by the first local modeler of the plurality of first local modelers, the central modeler executing on the stream processing system;

computing, by the central modeler, aggregated information using instances of partial aggregated information computed at least partially in parallel by the plurality of first local modelers; and

determining, by the central modeler, a plurality of parameters of a machine learning model using the aggregated information computed from the instances of partial aggregated information generated by the plurality of first local modelers, wherein the parameters of the machine learning model determine an assignment of events to respective sub-populations of events;

providing, by the central modeler, the plurality of parameters of the machine learning model to each of the first local modelers;

generating, by the central modeler, an updated routing strategy that assigns each event belonging to a particular sub-population of events according to the parameters of the machine learning model to a particular second local modeler of a plurality of second local modelers executing on the stream processing system;

providing, by the central modeler while the first local modelers continue to process events in the event stream, the updated routing strategy to a second router that routes events processed by the plurality of first local modelers, wherein the second router executes on the stream processing system;

processing, by the first local modelers, events in the event stream according to the parameters of the machine learning model to assign each event to a respective sub-population of events;

providing, by the first local modelers, the processed events to the second router; and

providing, by the second router, each processed event received from the first local modelers to a respective second local modeler according to a sub-population of events assigned by the first local modeler based on the updated routing strategy computed by the central modeler.