Friday, 17th September 2010, 2:00 pm (PDCC Meeting Room)
Speaker: Guozhang Wang (Cornell University)
Title: Behavioral Simulations in MapReduce
In many scientific domains, researchers are turning to large-scale behavioral simulations to better understand important real-world phenomena. While there has been a great deal of work on simulation tools from the high-performance computing community, behavioral simulations remain challenging to program and automatically scale in parallel environments.
In this talk, I will present a new behavioral simulation platform to give simulation scientists both easy programmability and scalability. To achieve ease of programming, the platform gives the simulation scientists a high-level language, called BRASIL (the Big Red Agent SImulation Language), which allows them to express simulations in a time-stepped pattern which we have developed for non-player characters in computer games but also very common in most behavioral simulation models, called State-Effect pattern. BRASIL has object-oriented features for programming simulations, but can be compiled to a data-flow representation for automatic parallelization and optimization.
For scalability, I will show you how to translate the state-effect pattern into MapReduce, so we can then scale simulations to large scenarios over a cluster of computers. For high performance, our platform also provides a special-purpose MapReduce engine optimized for behavioral simulations, called BRACE (the Big Red Agent-based Computation Engine), which leverage spatial locality to greatly reduce the communication between nodes. In our experiments our platform can achieve both scalability and single-node performance similar to that of hand-coded simulators on several realistic simulations.
This is joint work with the Cornell Database group and the Cornell School of Civil and Environmental Engineering.
Guozhang Wang is a third year PhD candidate at the Cornell University in the Department of Computer Science, working with Prof. Johannes Gehrke. His research interests are broadly in data management and cloud computing. He is specifically interested in bring data management techniques such as parallel query processing and query optimization to large scale data-driven applications. His current research focus is on building scalable behavioral simulation platforms.