14:00, Room N107 (Parc Club)
In many scientific domains, researchers are turning to large-scale behavioral simulations to better understand important real-world phenomena. These phenomena emerge as the result of a myriad interactions among large numbers of interdependent agents in a complex system, such as a transportation network or an ecological system. 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 show how database techniques can solve this dilemma, by offering simulation developers a programmable environment that automatically provides for scalability and durability.
My talk will be organized in two parts. In the first, more detailed part of my talk, I will discuss the design of BRACE, the Big Red Agent-based Computation Engine. BRACE leverages spatial locality to treat behavioral simulations as iterated spatial joins and greatly reduce communication between nodes in a cluster. While this set-at-a-time processing model can be very efficient, it can be much simpler for the domain scientist to program the behavior of a single agent. As a consequence, BRACE includes a high-level language called BRASIL (the Big Red Agent SImulation Language). BRASIL has object oriented features for programming simulations, but can be compiled to a database-style representation for automatic parallelization and optimization. In the second part of my talk, I will discuss techniques to provide efficient durability to behavioral simulations. The problem is challenging because these systems must sustain extremely high update rates, often hundreds of thousands of updates per second or more. We leverage the observation that simulations have frequent points of consistency to develop novel checkpoint recovery algorithms that trade additional space in main memory for significantly lower overhead and latency than existing methods. After presenting the database approach taken in BRACE, I will discuss directions of ongoing work on complex simulation models over cloud computing environments as well as directions of future work.
Marcos Vaz Salles is a postdoc at Cornell University. His research targets building novel data-driven systems that bring classic database benefits, such as scalability and ease of programming, to new domains. At Cornell, Marcos is currently working on data management techniques for computer games and behavioral simulations. During his PhD in the Systems Group at ETH Zurich, he investigated hybrid search and data integration architectures for personal dataspace management in the iMeMex project. Previously, Marcos obtained his MSc from PUC-Rio, Brazil, and his BSc from UNICAMP, Brazil.