Dr. Daniel Amir

Throughout his academic career, Daniel Amir has striven to design next-generation computer systems that are both practical and based in rigorous theoretical foundations. During his Ph.D. in Computer Science at Cornell University, Dr. Amir collaborated with world-class researchers in both network systems and theoretical computer science to develop Oblivious Reconfigurable Networks (ORNs), a novel network design paradigm for post-Moore’s Law datacenter networking.

Reconfigurable networks have become popular in datacenters due to the hardware advantages of optical switching. Unlike traditional electrical switches, these networks are not limited by slowdowns in transistor scaling. However, traditional reconfigurable network designs connect nodes based on current traffic demand, a cumbersome process which is too slow to support the short messages commonly found in datacenter traffic. ORNs reconfigure oblivious to the traffic demand, eliminating this limitation and allowing all traffic to be efficiently supported.

During his time at Cornell, Dr. Amir collaborated with leading theoreticians to prove the optimal performance tradeoffs achievable by ORNs. Using this theoretical foundation, he developed the first practical designs for datacenter-scale ORNs, which also achieve the ideal performance tradeoffs. He has validated his designs using detailed, datacenter-scale packet simulations.

For his postdoc in the Electrical and Computer Engineering Department of the Technion–Israel Institute of Technology, Dr. Amir hopes to study a previously unaddressed question in next-generation optical networks: where best to locate in‑network computation. As in-network computation has grown immensely in importance in today’s data centers, addressing this question is key to facilitating the growing deployment of optical switching technologies, including ORNs. He additionally hopes to further optimize ORNs for the specific traffic patterns found in datacenter applications, including those which interact with in-network computation, bringing ORNs even closer to practical deployment.