Datecenter Scheduler Design
Datacenter scheduling architectures, algorithms, and policies are well studied. However, the APIs the schedulers expose and the component features required to build an architecture less so. My student, Aratz, and I investigated the impact of scheduler API on the performance of Google, Azure, and BitBrains workloads (Lasa et al., 2024).
I investigated the feature different components, such as the scheduler, request router, and data manager, need to provide to enabled different scheduler architecture and mechanisms (Talluri et al., 2024).
With my colleague, Dante Niewenhuis, I investigated sustainability techniques for datacenters such (Niewenhuis et al., 2024). What size of batteries should I use? When should I run my workload?. This work is under submission.
