Technion Computer Science Colloquium Time+Place : Tuesday 28/12/2010 14:30 room 337-8 Taub Bld. Speaker : Danny Bickson Affiliation: Carnegie Mellon University Host : Johann Makowsky Title : GraphLab: Asynchronous Graph Computation in the Clouds and Beyond Abstract : As the amounts of collected data and computing power grows (multicore, GPUs, clusters, clouds), modern datasets no longer fit into one computing node. Efficient distributed/parallel algorithms for handling large scale data are required. Efficient solutions must take into account both system aspects and algorithmic aspects; solutions that ignore other aspects will not scale. In this talk, I will cover my work concerning both the theoretical and the practical aspects of parallel and distributed large scale computing on massive datasets. My work has focused on the following three elements: efficient algorithm design, compact yet accurate data representation and efficient large scale system building. Regarding efficient algorithm design, my work is mainly focused on the design of asynchronous iterative algorithms. Asynchronous algorithms are especially suitable for heterogeneous computing environments since they allow for flexible scheduling and thus efficient utilization of all computing resources. As for compact data models, I have worked on developing the theory and methods for handling heavy-tailed distributions, which naturally characterize many system workloads like network flows in a communication network. Having more accurate data representation allows for savings both in storage space and significant reduction of the required computation. As a case study for this approach, I will present the GraphLab framework which is a parallel programming abstraction targeted for sparse asynchronous graph algorithms. The Distributed GraphLab implementation is targeted to allow execution of large scale machine learning algorithms in parallel on multiple platforms, such as multicores, clusters and clouds. GraphLab provides a high level programming interface, allowing a rapid deployment of distributed algorithms. GraphLab is an open source project available for download on <http://graphlab.ml.cmu.edu/> Joint work with: Carlos Geustrin, Joseph Gonzalez, Yucheng Low, Aapo Kyrola, Joseph Bradely from Carnegie Mellon University and Joseph Hellerstein from Berkeley Refreshments served from 14:15 on, Lecture starts at 14:30 ---------------------------------------------------------- Visit our home page- <http://www.cs.technion.ac.il/~colloq> --------------------------------------------------------- Technion Math. Net (TECHMATH) Editor: Michael Cwikel <techm@math.technion.ac.il> Announcement from: Hadas Heier <heier@cs.technion.ac.il>