Ethan Berkley

Bio

Ethan Berkley is a full-time undergraduate research assistant at Cornell University's Computer Systems Laboratory (CSL), advised by Professor Mohammad Alian. His current work is investigating how ML accelerators can use remote memory within the scale up domain to augment on-chip HBM, mitigating data delivery bottlenecks. Furthermore, he is interested in exploring different ways to harvest overprovisioned compute / memory from accelerators running heterogeneous workloads, as well as how to use network-attached memory expanders to shift AI reliance away from HBM.

In the past, he has investigated how dynamic compute granularity in CPU workloads can affect resource efficiency, and created techniques to identify performant granularities that minimize resource allocations under mixed-QoS requirements.

Research Interests

• Systems for ML.
• Architectural Simulation.
• QoS-aware scheduling/allocation policies.
• Datacenter networking.