Parallel Computer Protocol Selection via Performance Fit

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current parallel computing systems face inefficiencies in selecting optimized protocols for collective operations, leading to suboptimal performance, resource utilization, and increased overhead due to the lack of accurate protocol selection methods.

Innovation Solution

A method for collective operation protocol selection in a parallel computer that iteratively evaluates and selects protocols based on predefined performance criteria, using a protocol performance function and a performance fit equation to determine the best protocol for executing collective operations, ensuring that the selected protocol meets minimum performance thresholds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a single protocol is used for all collective operations, then device complexity is reduced, but performance is suboptimal because the protocol may not be suited for specific operation parameters

Engineering Contradiction:
Improvecollective operation performanceVSAvoidprotocol selection mechanism complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system dynamically selects protocols based on operating parameters rather than using a static single protocol. The collective operation module evaluates parameters such as data size, node count, and operation type to adaptively choose the most suitable protocol from multiple available protocols, enabling the system to optimize performance for different operational contexts.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes protocol selection based on varying operating parameters. By monitoring parameters like message size, number of processes, and collective operation type, the system adjusts its protocol choice to match the current operational requirements, thereby improving performance without requiring complex manual configuration.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If multiple protocols are available for collective operations, then performance can be optimized for different scenarios, but device complexity increases due to protocol selection overhead

Engineering Contradiction:
Improvecollective operation throughputVSAvoidprotocol selection time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system performs preliminary evaluation of protocol performance using performance functions and fit equations before actual collective operations execute. By pre-assessing which protocols will perform best for given operating parameters, the system avoids time-consuming trial-and-error during runtime and makes informed protocol selections in advance.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses performance functions that evaluate how well protocols fit specific operating parameters, creating a feedback mechanism that guides protocol selection. This feedback-based approach allows the system to learn from performance evaluations and make increasingly accurate protocol choices, reducing selection overhead over time.

Inventive Principle:
Principle #23Feedback

3Productivity

If protocol selection is performed without performance criteria, then device complexity is minimized, but resource utilization becomes inefficient

Engineering Contradiction:
Improveresource utilization efficiencyVSAvoidperformance evaluation mechanism complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The collective operation module autonomously evaluates and selects protocols based on performance criteria without requiring external intervention. The system self-services by internally assessing operating parameters against performance functions and automatically choosing the most appropriate protocol, improving resource utilization without adding complex external control mechanisms.

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS9047091B2Collective operation protocol selection in a parallel computer
Publication Date: 2015.06.02 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • US9047091B2 patent drawing
  • US9047091B2 patent drawing
  • US9047091B2 patent drawing

AI summary

Collective operation protocol selection in a parallel computer that includes compute nodes may be carried out by calling a collective operation with operating parameters; selecting a protocol for executing the operation and executing the operation with the selected protocol. Selecting a protocol includes: iteratively, until a prospective protocol meets predetermined performance criteria: providing, to a protocol performance function for the prospective protocol, the operating parameters; determining whether the prospective protocol meets predefined performance criteria by evaluating a predefined performance fit equation, calculating a measure of performance of the protocol for the operating parameters; determining that the prospective protocol meets predetermined performance criteria and selecting the protocol for executing the operation only if the calculated measure of performance is greater than a predefined minimum performance threshold.