30 Jul 2026
A paper accepted at SEC 2026
Our paper “Occamy: Latency-Aware Scheduling with Controlled Degradation for Mixed-Criticality AI at the Edge”, in collaboration with Patient, has been accepted as a full paper at SEC 2026.
We extend the graceful-degradation methodology of Cerberus from latency relaxation to model-variant selection through the development of the Occamy global scheduling model, targeting mixed-criticality AI inference across distributed edge sites. Whereas Cerberus restores feasibility by relaxing the latency bounds of low-criticality applications, Occamy exploits the multiple model variants that AI applications expose, each trading inference accuracy for lower latency and resource consumption. Occamy predicts the P99 end-to-end latency of a candidate deployment and jointly selects a model variant, an instance count, and a user-to-site assignment for each application under strict per-application SLOs and edge capacity limits. Under resource scarcity, a priority-aware degradation algorithm switches low-criticality applications to lighter variants in controlled steps until a feasible placement is found, while preserving the nominal quality of high-criticality ones. This delivers graceful degradation along the accuracy dimension, keeping all applications operational with quality loss that is predictable and minimal.
Yinan will present the paper during the conference on October 13-16 2026 in Santa Clara, CA, United States.
03 Jul 2026
A paper accepted at SRDS 2026
The paper “Cerberus: Latency-Aware Scheduling with Graceful Degradation for Constrained Edge Infrastructures” has been accepted at SRDS 2026 as a full paper.
We extend the scheduling methodology of Scylla to handle multi-application concurrency under resource scarcity on edge clusters through the development of the Cerberus global scheduling model. Recognizing that edge resources are finite and may be insufficient to simultaneously satisfy all application SLOs under peak demand, Cerberus extends latency-aware scheduling with a graceful degradation mechanism that avoids the service disruptions caused by eviction-based strategies. Instead of dropping low-criticality applications, Cerberus introduces the concept of a Service Level Threshold (SLT), which bounds the maximum acceptable performance degradation for each application, and a priority-aware degradation algorithm that progressively relaxes the latency bounds of low-criticality applications in controlled steps until a feasible placement is found, while preserving strict SLO compliance for high-criticality ones. This ensures that all applications remain operational under resource scarcity, with degradation that is predictable, minimal, and proportional to application criticality.
Yinan will present the paper during the conference on September 22-24 2026 in Roma, Italy.
02 Jul 2026
Yinan presented the paper at IWQoS 2026
Yinan presented the paper “Scylla: Scheduling Multiple Latency-Sensitive Applications in the Edge-Cloud Continuum” at IWQoS 2026 on July 2nd 2026 in Istanbul, Turkey.
30 Apr 2026
A paper accepted at IWQoS 2026
The paper “Scylla: Scheduling Multiple Latency-Sensitive Applications in the Edge-Cloud Continuum” has been accepted at IWQoS 2026 as a regular paper in the Edge-Cloud Resource Allocation track.
We extend the scheduling methodology of LASSY to handle multi-application concurrency through the development of the Scylla global scheduling model. Recognizing that edge nodes are rarely dedicated to a single task, Scylla generalizes the single-application formulation to jointly schedule multiple co-located applications competing for shared resources across the edge-cloud continuum, mitigating the “Noisy Neighbor” effect through a holistic deployment plan that ensures all applications meet their respective SLOs without mutual performance interference. As the joint optimization of multiple applications causes the problem scale and computation time to grow rapidly, we further introduce the Layered Permutation Search (LPS) heuristic, which decomposes the global problem into a sequence of single-application problems and explores application orderings at controllable depth, achieving near-optimal placement plans within practical time limits even for large numbers of applications.
Yinan will present the paper during the conference during June 29 - July 2 2026 in Istanbul, Turkey.
13 Apr 2026
Maxime presented at VISSOFT/ICSME 2025
On April 13th, 2026, Maxime presented his works at MSR 2026.
The paper PoolinGH: Fast, Efficient, and Robust GitHub Repository Mining, approaches PoolinGH, a lightweight, open-source, easy-to-use library, aimed at supporting researchers.
