Application resource model composition from constituent components

• US 8,255,554 B2
• Filed: 05/14/2009
• Issued: 08/28/2012
• Est. Priority Date: 05/14/2009
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Andrade, Henrique, Gedik, Bugra, Parekh, Sujay Sunil, Wu, Kun-Lung, Zhang, Xiaolan
• Primary Examiner(s): Barqadle, Yasin
• Assistant Examiner(s): Chacko, Joe
• Application Number:
  US12/466,017
• Publication Number:
  US 20100292980A1
• Time in Patent Office: 1,202 Days
• Field of Search: 709/224, 717/133
• US Class Current: 709/231
• CPC Class Codes: G06F 9/5011 the resources being hardware resources other than CPUs, Servers and Terminals
• CPC Class Codes: H04L 43/091 Measuring contribution of individual network components to actual service level

Abstract

Techniques for composing an application resource model in a data stream processing system are disclosed. The application resource model may be used to understand what resources will be consumed by an application when executed by the data stream processing system. For example, a method for composing an application resource model for a data stream processing system comprises the following steps. One or more operator-level metrics are obtained from an execution of a data stream processing application in accordance with a first configuration. The application is executed by one or more nodes of the data stream processing system, and the application is comprised of one or more processing elements that are comprised of one or more operators. One or more operator-level resource functions are generated based on the obtained one or more operator-level metrics. A processing element-level resource function is generated based on the one or more generated operator-level resource functions. The processing element-level resource function represents an application resource model usable for predicting one or more characteristics of the application executed in accordance with a second configuration.

Claims

1. A method for composing an application resource model for a data stream processing system, the method comprising steps of:

obtaining one or more operator-level metrics from an execution of a data stream processing application in accordance with a first configuration, wherein the application is executed by one or more nodes of the data stream processing system, and wherein the application is comprised of one or more processing elements that are comprised of multiple operators, wherein two or more of the multiple operators are combined in a first combination to form a processing element, wherein the first combination of operators is in accordance with the first configuration, and wherein the operator-level metrics are obtained by learning a communication overhead function associated with communication between two or more processing elements;

wherein learning a communication overhead function comprises;

collecting calibration data from a source-sink benchmark to determine a maximum rate and a central processing unit (CPU) usage associated with executing the benchmark configuration; and

scaling the maximum rate and the CPU usage measured from the benchmark configuration according to a rate and a tuple size for which the overhead is to be computed;

generating one or more operator-level resource functions from the first combination of operators in accordance with the first configuration based on the obtained one or more operator-level metrics; and

generating a processing element-level resource function using the one or more generated operator-level resource functions to predict a model for the processing element formed by a second combination of two or more of the multiple operators, the processing element-level resource function representing an application resource model usable for predicting one or more characteristics of the application executed in accordance with a second configuration;

wherein the generating steps are executed on a computing system comprising a processor and a memory, wherein one or more software modules for implementing the generating steps are loaded from the memory and executed by the processor resulting in the generation of the application resource model.

2. The method of claim 1, wherein the first configuration of the application is a calibration configuration, and the second configuration of the application is an optimized configuration.

3. The method of claim 1, wherein the first configuration of the application comprises one or more processing elements executed on a first hardware configuration, and the second configuration of the application comprises one or more processing elements executed on a second hardware configuration.

4. The method of claim 1, wherein the one or more predicted characteristics comprise a resource consumption characteristic.

5. The method of claim 1, wherein the operator-level metrics are obtained by profiling a runtime environment associated with the execution of the application in the first configuration.

6. The method of claim 5, wherein the step of profiling the runtime environment further comprises:

learning a processing capacity of each node of the first configuration, wherein a node contains one or more central processing units (CPUs); and

learning a scaling factor for CPU usage caused by multiprocessing.

7. The method of claim 1, wherein the step of generating one or more operator-level resource functions further comprises:

adjusting for a runtime environment;

computing one or more fitting functions for each operator;

projecting a saturation point of one or more of the processing elements; and

deriving an operator-level resource function.

8. The method of claim 7, wherein the step of adjusting for a runtime environment further comprises:

normalizing a central processing unit (CPU) usage according to a specific node capacity; and

adjusting the CPU usage according to a specific environment from which the operator-level metrics were obtained.

9. The method of claim 1, wherein the step of generating the processing element-level resource function further comprises:

summing the resource functions of the operators of a given processing element;

summing a communication overhead function; and

adjusting for a runtime environment.

10. The method of claim 8, wherein the step of adjusting the CPU usage further comprises:

scaling the CPU usage according to a specific target node capacity; and

further adjusting the CPU usage to account for a specific environment and node on which the processing element is to be executed.

11. The method of claim 1, wherein the one or more operators encapsulate logic of an operation associated with a given processing element.

12. The method of claim 1, further comprising the step of using the processing element-level resource function representing the application resource model to predict the one or more characteristics of the application executed in accordance with a second configuration.

13. The method of claim 12, further comprising the step of deploying the application for subsequent execution depending on the one or more predicted characteristics.

14. Apparatus for composing an application resource model for a data stream processing system, the apparatus comprising:

a memory; and

a processor coupled to the memory and configured to;

(i) obtain one or more operator-level metrics from an execution of a data stream processing application in accordance with a first configuration, wherein the application is executed by one or more nodes of the data stream processing system, and wherein the application is comprised of one or more processing elements that are comprised of multiple operators, wherein two or more of the multiple operators are combined in a first combination to form a processing element, wherein the first combination of operators is in accordance with the first configuration, and wherein the operator-level metrics are obtained by learning a communication overhead function associated with communication between two or more processing elements;

wherein learning a communication overhead function comprises;

collecting calibration data from a source-sink benchmark to determine a maximum rate and a central processing unit (CPU) usage associated with executing the benchmark configuration; and

scaling the maximum rate and the CPU usage measured from the benchmark configuration according to a rate and a tuple size for which the overhead is to be computed;

(ii) generate one or more operator-level resource functions from the first combination of operators in accordance with the first configuration based on the obtained one or more operator-level metrics; and

(iii) generate a processing element-level resource function using the one or more generated operator-level resource functions to predict a model for the processing element formed by a second combination of two or more of the multiple operators, the processing element-level resource function representing an application resource model usable for predicting one or more characteristics of the application executed in accordance with a second configuration.

15. The apparatus of claim 14, wherein the first configuration of the application is a calibration configuration, and the second configuration of the application is an optimized configuration.

16. The apparatus of claim 14, wherein the first configuration of the application comprises one or more processing elements executed on a first hardware configuration, and the second configuration of the application comprises one or more processing elements executed on a second hardware configuration.

17. The apparatus of claim 14, wherein the one or more predicted characteristics comprise a resource consumption characteristic.

18. A computer program product comprising a non-transitory tangible computer readable recordable storage medium including computer usable program code for composing an application resource model for a data stream processing system, the computer program product comprising:

computer usable program code for obtaining one or more operator-level metrics from an execution of a data stream processing application in accordance with a first configuration, wherein the application is executed by one or more nodes of the data stream processing system, and wherein the application is comprised of one or more processing elements that are comprised of multiple operators, wherein two or more of the multiple operators are combined in a first combination to form a processing element, wherein the first combination of operators is in accordance with the first configuration, and wherein the operator-level metrics are obtained by learning a communication overhead function associated with communication between two or more processing elements;

wherein learning a communication overhead function comprises;

collecting calibration data from a source-sink benchmark to determine a maximum rate and a central processing unit (CPU) usage associated with executing the benchmark configuration; and

scaling the maximum rate and the CPU usage measured from the benchmark configuration according to a rate and a tuple size for which the overhead is to be computed;

computer usable program code for generating one or more operator-level resource functions from the first combination of operators in accordance with the first configuration based on the obtained one or more operator-level metrics; and

computer usable program code for generating a processing element-level resource function using the one or more generated operator-level resource functions to predict a model for the processing element formed by a second combination of two or more of the multiple operators, the processing element-level resource function representing an application resource model usable for predicting one or more characteristics of the application executed in accordance with a second configuration.

19. The computer program product of claim 18, wherein the first configuration of the application is a calibration configuration, and the second configuration of the application is an optimized configuration.

20. The computer program product of claim 18, wherein the first configuration of the application comprises one or more processing elements executed on a first hardware configuration, and the second configuration of the application comprises one or more processing elements executed on a second hardware configuration.