Arvindra Sehmi, Microsoft EMEA

Track:   Scalable Computing

Abstract: Application scalability and performance is hard to design and plan for. Distributed message processing systems should be designed using simple, efficient, and "mathematically" sound architectural building blocks resulting in more predicable and controllable systems in terms of scalability and performance. This talk will describe such architectures and give two case study examples to validate the approach. Developing this idea further we describe an "agent" model which uses "mental states" such as beliefs, desires, intentions and commitments to process messages. This agent architecture can be used to construct agile information processing machines capable of very high performance business transaction management. Then you will learn how a queuing network framework (which is itself a general purpose infrastructure for building fast, scalable message processing systems) was used to construct the agent and delve into the detailed design of a sample application for financial trade message matching. Key characteristics of this example scenario are: different kinds of messages represent different counterparties who trade amongst each other in collaboration with multiple third parties; messages need to be processed fast; differently; and very flexibly. These requirements are handled very well in the agent architecture and many ideas described throughout the talk apply equally well in other domains such as payments processing, order routing, funds transfer, telecoms network management, manufacturing process automation and supply & demand chain management.