Our paper titled "It Takes Two to Tango: Serverless Workflow Serving via Bilaterally Engaged Resource Adaptation" has been accepted to the 39th IEEE International Parallel and Distributed Processing Symposium. IPDPS is a premium conference in parallel and distributed systems and this year it will be held at Politecnico di Milano in Milan, Italy.
The paper addressed the information barrier between the application developer and the cloud provider for resource scaling in serverless computing systems. We propose Janus, a new serverless framework where application developers can generate resource-performance hints offline and supply them to the cloud provider to right size individual serverless functions in a workflow dynamically at runtime, achieving resource efficiency while guaranteeing end-to-end latency service-level objectives (SLOs) for the workflow.
This work is a collaboration between Paderborn University and the Huazhong University of Science and Technology.
