In this talk, Stefan Schmid presents the vision of self-adjusting networks: networks whose physical topology dynamically adapts according to the demand that it currently serves. Self-adjusting networks can appear in many different contexts, and a particularly interesting emerging opportunity arises in datacenters, especially when machine learning applications generate highly structured communication patterns, which can be exploited by self-adjusting networks using emerging reconfigurable optical technologies.

Stefan will survey enabling technologies and discuss theoretical foundations, which are rooted in information theory and in algorithms. The talk will be based on our recent Communications of the ACM article https://cacm.acm.org/research/revolutionizing-datacenter-networks-via-reconfigurable-topologies/.