Simon Kassing, Vojislav Dukic, Ce Zhang, Ankit Singla

Explicit bounds on degradable traffic can improve multiplexing while protecting latency-sensitive workloads.

Vojislav Dukic

Workload knowledge can guide network design and scheduling to improve performance while reducing cloud infrastructure cost.

Vojislav Dukic, Rodrigo Bruno, Ankit Singla, Gustavo Alonso

Sharing runtimes across concurrent serverless invocations reduces memory use and cold starts without sacrificing performance.

Vojislav Dukic, Ginni Khanna, Christos Gkantsidis, Thomas Karagiannis, Francesca Parmigiani, Ankit Singla, Mark Filer, Jeffrey L. Cox, Anna Ptasznik, Nick Harland, Winston Saunders, Christian Belady

Optical circuit switching can make flexible, low-latency regional data-center networks economically viable.

Vojislav Dukic, Ankit Singla

A lightweight feedback interface can help cloud providers discover better resource configurations for tenant workloads.

Vojislav Dukic, Sangeetha Abdu Jyothi, Bojan Karlas, Muhsen Owaida, Ce Zhang, Ankit Singla

Partial knowledge of flow sizes is practical and useful for improving data-center scheduling.

Vojislav Dukic, Sangeetha Abdu Jyothi, Bojan Karlas, Muhsen Owaida, Ce Zhang, Ankit Singla

Workload signals can make network scheduling more effective, even when applications cannot provide flow sizes directly.

Vojislav Dukic, Ivan Luković, Matej Črepinšek, Tomaž Kosar, Marjan Mernik

Application-level traceability makes complex information systems easier to evolve and verify.