Compute intensive stream processing with context data routing

Patent metadata

Field	Value
Publication number	US-11343156-B2
Application number	US-201514710152-A
Country	US
Kind code	B2
Filing date	May 12, 2015
Priority date	Sep 8, 2014
Publication date	May 24, 2022
Grant date	May 24, 2022

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for routing events of an event stream in a stream processing system. One of the methods includes receiving, by a router, an event stream of events; identifying, for each event, by the router, a respective partition of context data that includes context data related to the event and providing the event to a respective local modeler that stores the partition of context data identified for the event in operational memory of the local modeler; processing, by each local modeler, events received from the router and aggregating information associated with each event to generate aggregated information; providing, by one or more of the local modelers, to a central modeler, the respective aggregated information; and determining, by the central modeler, a plurality of parameters of a machine learning model using the received aggregated information.

First claim

What is claimed is:

1. A method comprising:
receiving, by a router, an event stream of events, wherein the router executes on a stream processing system comprising one or more computers, wherein the stream processing system includes a first plurality of local modelers, wherein the stream processing system partitions context data into a plurality of partitions of context data stored across the first plurality of local modelers, and wherein each local modeler of the first plurality of local modelers maintains a different respective partition of context data in operational memory of the local modeler;
identifying, for each event in the event stream, by the router, a respective partition of context data that includes first context data related to the event;
providing each event in the event stream, by the router, to a respective local modeler of the first plurality of local modelers identified as storing the respective partition of context data that includes first context data related to the event;
processing, by each local modeler of the first plurality of local modelers, events received from the router using the first context data in the respective partition of first context data by performing a first plurality of actions, the actions including aggregating information associated with each event received by the local modeler to generate aggregated information;
providing, by one or more of the local modelers, to a first central modeler executing on the system of one or more computers, the respective aggregated information generated by the one or more of the local modelers; and
determining, by the first central modeler, a plurality of parameters of a first machine learning model using the aggregated information received by the first central modeler, wherein each local modeler is implemented as a respective system of one or more computers that is separate from other systems of one or more computers implementing the other local modelers of the first plurality of local modelers, wherein the router is an ingestion system separate from the local modelers of the first plurality of local modelers, and wherein the central modeler is implemented as a respective system of one or more computers that is separate from the systems implementing the local modelers of the first plurality of local modelers and separate from the router.

2. The method of claim 1 , wherein a first local modeler and a second local modeler of the first plurality of local modelers each store a same partition of first context data in respective operational memory of the first local modeler and the second local modeler.

3. The method of claim 1 , further comprising:
before identifying a partition of context data, determining that an action of the first plurality of actions requests the first context data.

4. The method of claim 1 , wherein:
for each local modeler of the first plurality of local modelers, instructions implementing the first plurality of actions and the respective partition of first context data are stored in operational memory in a respective operating system process; and
each local modeler processes, in the respective operating system process, events received by the local modeler by performing the first plurality of actions.

5. The method of claim 1 , further comprising:
providing, by the first central modeler, the parameters to one or more of the first plurality of local modelers implemented on respective separate systems of one or more computers for storage on the respective systems.

6. The method of claim 5 , further comprising:
processing, by each of the one or more of the first plurality of local modelers applying the first machine learning model, the events received by the local modeler to perform scoring of the event stream; and
aggregating, by each of the one or more of the first plurality of local modelers, information associated with each event received by the local modeler to generate updated aggregated information.

7. The method of claim 1 , wherein identifying a partition of context data that includes context data related to an event comprises:
obtaining information identifying a first information type included in the event;
obtaining data included in the event that is associated with the first information type; and
performing a hashing process using the obtained data to identify the partition.

8. The method of claim 1 , wherein the router is a first router, the method further comprising:
providing, by the first plurality of local modelers, processed events to a second router different from the first router;
identifying, by the second router, for each processed event, a respective partition of second context data that includes second context data related to the processed event; and
providing each processed event, by the second router, to a respective local modeler of a second plurality of local modelers identified as storing a partition of second context data that includes second context data related to the processed event.

9. The method of claim 8 , further comprising:
processing, by each local modeler in the second plurality of local modelers, events received from the second router using the second context data in the respective partition of second context data by performing a second plurality of actions.

10. The method of claim 1 , wherein the router is a first router, the method further comprising:
providing, by the first plurality of local modelers, processed events to a second router different from the first router, wherein the processed events include further information added by local modelers in the first plurality of local modelers; and
identifying, by the second router, for each processed event, using the further information added to the processed event, a respective local modeler of a second plurality of local modelers of the stream processing system, wherein the local modelers in the second plurality of local modelers execute on the system of one or more computers.

11. The method of claim 10 , wherein the further information is a label classifying the respective event.

12. The method of claim 1 , comprising:
providing, by the one or more local modelers of the first plurality of local modelers, to a second central modeler executing on the system, aggregated information generated by the one or more local modelers of the first plurality of local modelers; and
determining, by the second central modeler, parameters of a second machine learning model using the aggregated information received by the second central modeler.

13. The method of claim 12 , comprising:
receiving, by the first central modeler, data specifying a first learning program for determining parameters of the first machine learning model from aggregated information received by the first central modeler; and
receiving, by the second central modeler, data specifying a third learning program for determining parameters of a third machine learning model from aggregated information received by the second central modeler.

14. The method of claim 1 , wherein:
context data is pre-existing data related to an event that does not occur within data attributes of the event itself.

15. The method of claim 1 , wherein:
for each local modeler, instructions implementing the first plurality of actions and the respective partition of first context data are stored in operational memory in a respective operating system process;
for each local modeler, common context data is stored in operational memory in each respective operating system process; and
each local modeler processes, in the respective operating system process, events received by th

Assignees

• Pivotal Software Inc

Inventors

• Brand Michael
• Adams Lyndon John
• Brown David Russell
• Ng Kee Siong

Classifications

• G06N5/04

  Inference or reasoning models · CPC title

• H04L41/16 Primary

  using machine learning or artificial intelligence · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06N5/02

  Knowledge representation; Symbolic representation · CPC title

• H04L45/08

  Learning-based routing, e.g. using neural networks or artificial intelligence · CPC title