Profile-driven data stream processing

• US 8,601,458 B2
• Filed: 05/14/2009
• Issued: 12/03/2013
• 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, Wu, Kun-Lung
• Primary Examiner(s): Zhen, Li B
• Assistant Examiner(s): WHEATON, BRADFORD F
• Application Number:
  US12/465,757
• Publication Number:
  US 20100293535A1
• Time in Patent Office: 1,664 Days
• Field of Search: 717/155, 717/156
• US Class Current: 717/156
• CPC Class Codes: G06F 8/433 Dependency analysis; Data or control flow analysis
• CPC Class Codes: G06F 8/44 Encoding

Abstract

Techniques for compiling a data stream processing application are provided. The techniques include receiving, by a compiler executing on a computer system, source code for a data stream processing application, wherein the source code comprises source code for a plurality of operators, each of which performs a data processing function, determining, by the compiler, one or more characteristics of operators within the data stream processing application, grouping, by the compiler, the operators into one or more execution containers based on the one or more characteristics, and compiling, by the compiler, the source code for the data stream processing application into executable code, wherein the executable code comprises a plurality of execution units, wherein each execution unit contains one or more of the operators, wherein each operator is assigned to an execution unit based on the grouping, and wherein each execution unit is to be executed in a partition.

Claims

1. A method for compiling a data stream processing application, wherein the method comprises:

receiving, by a compiler executing on a computer system, source code for a data stream processing application, wherein the source code comprises source code for a plurality of operators, each of which performs a data processing function;

determining, by the compiler, one or more characteristics of the plurality of operators within the data stream processing application, wherein said determining comprises;

injecting profiling code into an instrumented version of the data stream processing application;

running the data stream processing application under a sample workload;

using the profiling code to collect one or more computation and communication characteristics of the plurality of operators within the data stream processing application; and

processing the one or more collected computation and communication characteristics to compute (i), for each of the plurality of operators, an average amount of demanded processing resources and, (ii) for each port of the plurality of operators, a mean data rate;

grouping, by the compiler, the plurality of operators into one or more execution containers based on the average amount of demanded processing resources and the mean data rate computations; and

compiling, by the compiler, the source code for the data stream processing application into executable code, wherein the executable code comprises a plurality of execution units, wherein each execution unit contains one or more of the plurality of operators, wherein each operator is assigned to an execution unit based on the grouping, and wherein each execution unit is to be executed in a partition.

2. The method of claim 1, wherein:

compiling the source code for the data stream processing application into executable code comprises fusing the plurality of operators until overhead of one or more tuple transfers is a pre-determined fraction of total cost of executing all operator logic within an execution container.

3. The method of claim 1, further comprising creating one or more containers that map a data flow graph of one or more data stream operators into a set of one or more executables that are deployable across a set of nodes.

4. The method of claim 3, wherein creating one or more containers comprises creating one or more containers that fuse one or more operators such that one or more stream connections between the one or more operators within the same container are reduced to one or more function calls.

5. The method of claim 1, further comprising re-targeting the application to a different hardware setup by compiling the application to match a physical flow graph.

6. The method of claim 1, further comprising outputting compiled source code, wherein the compiled source code comprises one or more operators fused into one or more partitions.

7. The method of claim 1, wherein the method is run on a system, wherein the system comprises one or more distinct software modules, each of the one or more distinct software modules being embodied on a tangible computer-readable recordable storage medium, and wherein the one or more distinct software modules comprise a profiling module, an optimizer module, a code generation module and a compiler module executing on a hardware processor.

8. A computer program product comprising a non-transitory tangible computer readable recordable storage device including computer useable program code for compiling a data stream processing application, wherein the computer usable program code comprises one or more distinct software modules, the computer program product including:

computer useable program code for receiving, by a compiler executing on a computer system, source code for a data stream processing application, wherein the source code comprises source code for a plurality of operators, each of which performs a data processing function;

computer useable program code for determining, by the compiler, one or more characteristics of the plurality of operators within the data stream processing application, wherein said determining comprises;

injecting profiling code into an instrumented version of the data stream processing application;

running the data stream processing application under a sample workload;

using the profiling code to collect one or more computation and communication characteristics of the plurality of operators within the data stream processing application; and

processing the one or more collected computation and communication characteristics to compute (i), for each of the plurality of operators, an average amount of demanded processing resources and, (ii) for each port of the plurality of operators, a mean data rate;

computer useable program code for grouping, by the compiler, the plurality of operators into one or more execution containers based on the average amount of demanded processing resources and the mean data rate computations; and

computer useable program code for compiling, by the compiler, the source code for the data stream processing application into executable code, wherein the executable code comprises a plurality of execution units, wherein each execution unit contains one or more of the plurality of operators, wherein each operator is assigned to an execution unit based on the grouping, and wherein each execution unit is to be executed in a partition.

9. The computer program product of claim 8, wherein the computer useable program code for compiling the source code for the data stream processing application into executable code comprises computer useable program code for fusing the plurality of operators until overhead of one or more tuple transfers is a pre-determined fraction of total cost of executing all operator logic within an execution container.

10. The computer program product of claim 8, further comprising:

computer useable program code for outputting compiled source code, wherein the compiled source code comprises one or more operators fused into one or more partitions.

11. The computer program product of claim 8, further comprising:

computer useable program code for creating one or more containers that map a data flow graph of one or more data stream operators into a set of one or more executables that are deployable across a set of nodes.

12. The computer program product of claim 11, wherein the computer useable program code for creating one or more containers comprises computer useable program code for creating one or more containers that fuse one or more operators such that one or more stream connections between the one or more operators within the same container are reduced to one or more function calls.

13. The computer program product of claim 8, wherein the one or more distinct software modules comprise a profiling module, an optimizer module, a code generation module and a compiler module executing on a hardware processor.

14. A system for compiling a data stream processing application, comprising:

a memory; and

at least one processor coupled to the memory and operative to;

receive, by a compiler executing on a computer system, source code for a data stream processing application, wherein the source code comprises source code for a plurality of operators, each of which performs a data processing function;

determine, by the compiler, one or more characteristics of the plurality of operators within the data stream processing application, wherein said determining comprises;

injecting profiling code into an instrumented version of the data stream processing application;

running the data stream processing application under a sample workload;

using the profiling code to collect one or more computation and communication characteristics of the plurality of operators within the data stream processing application; and

processing the one or more collected computation and communication characteristics to compute (i), for each of the plurality of operators, an average amount of demanded processing resources and, (ii) for each port of the plurality of operators, a mean data rate;

group, by the compiler, the plurality of operators into one or more execution containers based on the average amount of demanded processing resources and the mean data rate computations; and

compile, by the compiler, the source code for the data stream processing application into executable code, wherein the executable code comprises a plurality of execution units, wherein each execution unit contains one or more of the plurality of operators, wherein each operator is assigned to an execution unit based on the grouping, and wherein each execution unit is to be executed in a partition.

15. The system of claim 14, wherein the at least one processor coupled to the memory operative to compile the source code for the data stream processing application into executable code is further operative to fuse the plurality of operators until overhead of one or more tuple transfers is a pre-determined fraction of total cost of executing all operator logic within an execution container.

16. The system of claim 14, wherein the at least one processor coupled to the memory is further operative to:

output compiled source code, wherein the compiled source code comprises one or more operators fused into one or more partitions.

17. The system of claim 14, wherein the at least one processor coupled to the memory is further operative to:

create one or more containers that map a data flow graph of one or more data stream operators into a set of one or more executables that are deployable across a set of nodes.

18. The system of claim 17, wherein the at least one processor coupled to the memory operative to create one or more containers is further operative to create one or more containers that fuse one or more operators such that one or more stream connections between the one or more operators within the same container are reduced to one or more function calls.

19. The system of claim 16, further comprising a tangible computer-readable recordable storage medium having one or more distinct software modules embodied thereon, the one or more distinct software modules comprise a profiling module, an optimizer module, a code generation module and a compiler module executing on a hardware processor.