Scylla: Scheduling Multiple Latency-Sensitive Applications in the Edge-Cloud Continuum
(2026) 2026 IEEE/ACM 34rd International Symposium on Quality of Service (IWQoS) — Location: Istanbul, Turkey (29.June.2026)
Files
scyall_iwqos_approved.pdf
Open Access - Adobe PDF
- 1.57 MB
Details
- Authors
- Abstract
- (en) Cloud-hosted applications face significant latency challenges, particularly for users located far from data centers. Edge computing mitigates these issues by geographically distributing processing. However, the limited resources of edge sites must be allocated carefully to different applications to avoid contention that would outweigh latency benefits. We present Scylla, a multi-application, latency-aware scheduler for edge/cloud environments. Scylla accurately predicts the relationship between resources and tail latency using a queuing model faithful to the load-balancing mechanisms used in Kubernetes. It computes a global scheduling plan for multiple applications that ensures the respect of tail latency objectives, while minimizing resource consumption. We evaluate Scylla on real-world infrastructure using Kubernetes and several latency-sensitive applications, including edge AI. Our results show that Scylla effectively predicts tail latencies and makes scheduling decisions that utilize just the right amount of resources with low computational overhead, in contrast to state-of-the-art schedulers used in the industry or proposed in earlier research.
- Affiliations
Citations
Cao, Y., Rivière, E., & Sadre, R. (2026, July 3). Scylla: Scheduling Multiple Latency-Sensitive Applications in the Edge-Cloud Continuum. 2026 IEEE/ACM 34rd International Symposium on Quality of Service (IWQoS), Istanbul, Turkey. https://doi.org/10.1109/ccgrid64434.2025.00029