DYNAMIC WORKLOAD BALANCING FOR REAL-TIME STREAM DATA ANALYTICS

• US 20160182588A1
• Filed: 12/23/2014
• Published: 06/23/2016
• Est. Priority Date: 12/23/2014
• Status: Active Grant

• Current Assignee: Teradata US, Inc. (Teradata Corporation)
• Original Assignee: Teradata US, Inc. (Teradata Corporation)
• Inventors: Brealey, Elizabeth Jacqueline Stone, Luo, Congnan, Yuan, Heng, Milby, Gregory Howard
• Granted Patent: US 9,456,014 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24568 Data stream processing; Continuous queries
• CPC Class Codes: G06F 16/284 Relational databases
• CPC Class Codes: H04L 65/611 for multicast or broadcast systems for broadcast or conference H04L12/18; arrangements for broadcast or distribution combined with broadcast H04H20/00; arrangements for broadcast applications with a direct linkage to broadcast information or to broadcast space-time H04H60/00; selective distribution of broadcast services, e.g. multimedia broadcast multicast service [MBMS], H04W4/06

Abstract

Router threads read real-time stream data as that data is received at processing nodes along a processing path for the data. The data is initially parsed into workloads. Each router thread maintains real-time analytics for the workloads and reports deviations in the analytics to a workload coordinator. The workload coordinator dynamically and in real time provides updated data distribution maps based on the reported deviations, and the router threads use the data distribution maps to determine a next processing unit to forward the workloads, where the next processing unit includes an instance of a next processing node in the processing path for the workload. The next processing node performs additional processing on the workloads along the processing path.

Claims

1. A method, comprising:

reading, by a processor, real-time stream data as workloads, each workload including one or more data items from the real-time stream data;

determining, by the processor, any deviation in content of the workloads relative to previous content of previous workloads;

reporting, by the processor, when a particular deviation exceeds a threshold value to a workload coordinator; and

receiving, by the processor, a data distribution map dynamically broadcasted by the workload coordinator, the data distribution map indicates a processing unit to forward the workloads along a distribution path for the workloads where an instance of a next processing node for the workloads can process the workload.

2. The method of claim 1 further comprising, processing other independent instances of the method in parallel with the method on the other processing units.

3. The method of claim 1 further comprising, processing the method in a parallel processing Relational Database Management System (RDBMS).

4. The method of claim 1, wherein reading further includes aggregating selective data items parsed from the real-time stream data into groups, each group representing a particular one of the workloads.

5. The method of claim 2, wherein aggregating further includes selecting a specification for the real-time stream data that defines each of the groups based on a real-time stream data type assigned to the real-time stream data.

6. The method of claim 1, wherein reading further includes filtering a portion of the content associated with the workloads.

7. The method of claim 1, wherein reading further includes augmenting a portion of the content associated with the workload with other content.

8. The method of claim 1, wherein reading further includes buffering each of the workloads as received in a First-In First-Out (FIFO) processing queue.

9. The method of claim 1, wherein determining further includes checking for the deviation at preconfigured intervals of time.

10. The method of claim 1, wherein reporting further includes obtaining the threshold value as a predefined percentage value assigned to a real-time stream data type associated with the real-time stream data.

11. A method, comprising:

identifying, by a processor, a total number of processing units available to process workloads of real-time stream data;

obtaining, by the processor, a data delivery path for the workloads, each node of the data delivery path representing a type of processing performed on the workloads;

developing, by the processor, a data distribution map for each node that instructs each node as to which processing unit to forward the workloads where a next node along the delivery path can next process the workloads;

broadcasting, by the processor, the data distribution maps to the nodes;

receiving, by the processor, real-time analytics from each of the nodes when the workloads are processed through the data delivery path; and

dynamically updating, by the processor, the data distribution maps for the nodes in response to the real-time analytics and rebroadcasting the updated data distribution maps to the nodes.

12. The method of claim 11 further comprising, dynamically maintaining, by the processor, a global data distribution map that represents an aggregated view of the data distribution maps.

13. The method of claim 12, wherein dynamically maintaining further includes updating the global data distribution map in response to updates to one or more of the updated data distribution maps.

14. The method of claim 11 further comprising, processing the method in a parallel processing Relational Database Management System (RDBMS), wherein each processing unit enabled to process one or more types of processing associated with one or more the nodes in parallel with remaining processing units.

15. The method of claim 11, wherein developing further includes generating the data distribution maps based on a total number of workload average data sizes and workload types in view of the total number of processing units.

16. The method of claim 11, wherein receiving further includes obtaining the real-time analytics as deviations detected by each of the nodes in one or more of:

content associated with the workloads and a frequency of the workloads appearing at the nodes.

17. The method of claim 11, wherein dynamically updating further includes processing an algorithm that load balances workload processing across the processing units and the load balanced workload processing reflected in the updated data distribution maps.

18. A system, comprising:

processors arranged in a parallel processing Relational Database Management System (RDBMS); and

a dynamic workload balancer configured to;

i) execute on one or more of the processors, ii) dynamically develops a global data distribution map that defines a next processing unit for a workload associated with real-time stream data processing along a distribution path of a network, wherein the distribution path includes a plurality of processing nodes and each node representing a type of processing performed on the workload iii) dynamically updates the global data distribution map in response to real-time deviations detected in real-time stream data analytics reported by the processing nodes while processing the workload, and iv) dynamically broadcasts the global data distribution map to router threads processing on each node along the distribution path to rebalance the real-time stream data processing in real time.

19. The system of claim 18, wherein each processing unit is processed on one or more of the processors.

20. The method of claim 19, each processor is configured to process a plurality of independent processing node instances and each processing node instance is processed on one or more of the processing units in parallel with remaining processing node instances appearing on other ones of the processors.