IPC Configuration Session Memory Usage/Size Challenges with IBM Models - An Overview

Bill Dorow is a Senior Software Engineer with IBM in Rochester, Minnesota, where he has spent fifteen years in product and solution configuration, with deep experience in various configuration technologies (Trilogy, Selectica, and now SAP), and responsibilities ranging from modeling to application development to performance management to general solution architecture. Bill is currently involved in IBM's Blue Harmony Project, a very large SAP implementation program, where he fills the role of Configurator Architect. Bill earned his Bachelor of Science in Computer Science at Northern Michigan University in 1994, and is currently pursuing his Master of Science in Computer Science at the University of Minnesota, where his work will be completed in 2010.

Doug Ewing is a Senior Software Engineer with IBM, currently working from his home office in Fairport (Rochester), New York.  He has worked eleven years in product and solution configuration, utilizing both Trilogy's and SAP's configuration technologies.  His responsibilities have included modeling, architecture, performance management, and technical lead. Doug is currently involved in IBM's Blue Harmony Project, a very large SAP implementation program, where he fills the role of Modeling Architect. Doug earned his BASc in Mechanical Engineering at the University of Waterloo, Canada, in 1992.  He worked as a Trilogy senior modeling consultant from 1999 through 2003, when he joined IBM.

IBM's solution/system level product lines presents significant complexity in configuration. Orders may include racks, servers, storage, software, and services, while also allowing customization of placement, connections and other logical associations. Some configurations may include multiple servers, spread across multiple racks.

Given the size and complexity of these configurations, the resulting size of the IPC configuration sessions has been a problem area and much effort has been dedicated to memory footprint and performance analysis of the SAP IPC sessions including the models and the underlying engine.

 In this presentation we will give an overview of our approach to analysis, what we have learned and what we have left to do. In addition, we will show how constraints affect memory, reviewing what we have learned about the IPC engine such as the use of discrimination nets (Rete trees) in the pattern matching system (PMS), and the handling of facts in the truth maintenance system (TMS). The discussion will relate example dependencies to memory consumption and introduce recent optimizations in dependency compilation (OSS notes).

This will be a high level overview which will be expanded on in a break out session.