CISCO

ANTlab has been awarded with a Cisco Research Grant that aims at establishing long term partnerships with universities to foster research and innovation.

The research activity focusses on routing in datacenter networks. Routing is the process of selecting the best path for packets to travel across a network. This is a core functionality in today’s datacenters, as they are built with massive path redundancy. The rise of Artificial Intellingence and Machine Learning (AI/ML) has brought new challenges when designing routing schemes.

The problem is that AI/ML workloads are:

• very sensitive to suboptimal path selection;
• bursty and characterized by on/off patterns;
• synchronous (many flows showing up at the same time).

All these characteristics make it really hard to build an optimal routing scheme. In this project, the insight is that we can leverage recent advances in network programmability (i.e., IPv6 segment routing) to architect a new routing solution that adapts well to AI/ML workloads and that is able to optimize for
performance (i.e., throughput and latency).