December 2024 - Present
Integrated SeBS (Serverless Benchmarks Suite) workloads (e.g., video processing, graph algorithms) to improve model accuracy and characterize multi-application interference. Expanding GraphSAGE-based resource contention modeling, which generated 32-dim embeddings from system resource graphs for gzip, bzip2, and xz. Refining heterogeneous graph modeling, addressing previous limitations of homogenized node attributes. Creating an efficient representation of heterogeneous node features to improve contention prediction accuracy. Currently developing a dynamic microservice scheduler through Bayesian optimization and graph application embeddings to outperform standard heuristic-based schedulers like Kubernetes.
TBD