27 Parallel Performance Patterns

Mathew Schemenaur, Perot Systems Corporation

Waterfront Centre Hotel
Malaspina Room

This tutorial presents a family of design patterns that improve software performance with an emphasis on parallel performance. These design patterns are categorized and presented using the UML notation. The parallel work patterns are organized into the following categories: Work Model, Eliminate Contention, Increase Parallel Section, and Resource Conservation. The essence of software design is making trade-off between competing objectives. Performance objectives are often at the center of this trade-off controversy. Designing a software unit for performance always adds complexity to the raw functionality. As a result, the performance objective is often competing with needs for design simplicity and flexibility. The consistent reuse of performance patterns will counter the added complexity. Consistency within a system will give the developer a sense of familiarity, even when examining unexplored software units.

Participants will acquire understanding of several advanced performance patterns. The experience will enable designers to clearly attack the performance objectives without loosing sight of the objectives for simplicity and flexibility. Participants can avoid re-inventing existing performance solutions and improve the clarity of their designs.

Attendee Background: The tutorial is aimed at software professionals wanting to learn design patterns that support performance issues. Participants should have knowledge of object-oriented principles and operating systems concepts like threads and concurrency.

Mathew Schemenaur has been designing object-oriented systems for over 10 years. During that time he has worked on several domain specific frameworks. For three years, he worked on a high performance object-oriented system for a major airline company. During this period, he used design patterns to capture and communicate patterns for parallel performance.