Abstract: Large-scale IT services are now increasingly run in memory due to tight application latency demands. With services centered mostly around data, datacenter owners often integrate as much DRAM into a single blade as technology allows, and use low-latency high-bandwidth network fabrics to aggregate near-neighbor DRAM into large memory pools. Because DRAM now accounts for a substantial fraction of both server capital and operation costs, datacenter designers are increasingly customizing server hardware, software and infrastructure for online services around memory to reduce costs. In this talk, I will first motivate specialized server design for in-memory computing and then present promising avenues to explore specialization to improve server efficiency.
Bio: Babak is Professor in the School of Computer and Communication Sciences and the founding director of the EcoCloud research center investigating data-centric cloud technologies at EPFL. He has made numerous contributions to multiprocessor server design including an architecture underlying the Sun (now Oracle) WildFire/WildCat servers, memory system technologies incorporated in IBM BlueGene/P and Q, and server evaluation methodologies in use by AMD and HP. His latest proposal for server processor designs for scale-out services has laid the foundation for specialized processors such as Cavium ThunderX. He is a recipient of an NSF CAREER award, IBM Faculty Partnership Awards, and an Alfred P. Sloan Research Fellowship. He is a fellow of IEEE.