This colloquium can be seen in person at the Grainger Auditorium, 1002 ECE Bldg., and is also being streamed and recorded on Echo 360, https://echo360.org.The live streaming will be available only to students enrolled in ECE 500. Others are advised to attend the colloquium in person. The recording will be available to the public.
Datacenters are growing at unprecedented speeds building a foundation for global IT services, cost-effective containerized apps and novel paradigms including microservices and serverless computing. At the same time, we are entering a new era in computing where scalability no longer comes from higher density in silicon fabrication processes. Now, more than ever server designers are in search of new avenues to bridge the gap between higher demands for scalability and the diminishing returns in server density. In this talk, I will go over the basic anatomy of system hardware and software in a modern server blade which is primarily derived from the CPU-centric desktop PC of the 80s. I will then present opportunities for a clean slate design of servers based on integration, specialization and approximation as three pillars to enable server scalability in the post-Moore era.
Babak Falsafi is a Professor and the founding director of EcoCloud at EPFL. His contributions to computer systems include the first NUMA multiprocessors built by Sun Microsystems (WildFire/WildCat), memory streaming integrated in IBM BlueGene (temporal) and ARM cores (spatial), and performance evaluation methodologies in use by AMD, HP and Google PerfKit. He has shown that memory consistency models are neither necessary nor sufficient to achieve high performance in servers. These results led to fence speculation in modern CPUs. His work on workload-optimized server processors laid the foundation for the first generation of Cavium ARM server CPUs, ThunderX. He is a recipient of an Alfred P. Sloan Research Fellowship, and a fellow of ACM and IEEE.