Average Response Time Optimization for Service-Chain Routing in Computing Power Networks via Robust Queueing Network Analysis
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 24494-24511
DOI: 10.1109/tmc.2026.3735391 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Wenzheng Wang und 5 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Gefördert von: National Key Research and Development Program of China, Institute of Information & Communications Technology Planning & Evaluation, Korea government.
Abstract (Originalsprache)
Computing power networks (CPNs) have emerged as a promising paradigm for coordinating edge clouds, central clouds, and network infrastructures to support low-latency, high-quality distributed compute-intensive services. We consider a CPN service-chain routing system in which requests traverse multiple services in a prescribed order and are routed among candidate service instances at each stage. Our goal is to develop an explicit method that directly evaluates and optimizes the average response time (ART) of a service-routing policy using the multi-time-scale variability of request arrivals. Existing approaches either optimize surrogate metrics such as path length and queue load or rely on idealized independent decompositions of product-form queueing networks, which can lead to substantial deviations from the true ART. We propose a robust-queueing-network-based service-chain routing model (RCRM) that uses routing probabilities as decision variables and jointly represents service processing and multi-hop forwarding in a fine-grained, class-aware, non-product-form queueing network. RCRM represents the class-dependent coupled propagation of multi-time-scale external-arrival variability through service-side request queues and network-side bit-level forwarding queues. This representation supports more reliable evaluation of the delay components of ART. We then construct a differentiable robust queueing latency evaluator (RQLE) and develop a heuristically guided automatic-differentiation delay optimization algorithm (HADO) that exploits the white-box structure of RCRM to efficiently search for low-latency service-chain routing policies. We implement a discrete-event CPN simulator for validation. The results show that RQLE achieves average estimation errors of $3.1\%$ 3 . 1 % – $12.0\%$ 12 . 0 % , while HADO reduces ART by $11.0\%$ 11 . 0 % – $32.4\%$ 32 . 4 % overall relative to the best baselines.
Förderung
- National Key Research and Development Program of China
- Institute of Information & Communications Technology Planning & Evaluation
- Korea government
- National Natural Science Foundation of China
- China Postdoctoral Science Foundation
Zitieren
APA
Wang, W., Kim, D. I., Xie, R., Tang, Q., Xie, G., & Huang, T. (2026). Average Response Time Optimization for Service-Chain Routing in Computing Power Networks via Robust Queueing Network Analysis. IEEE Transactions on Mobile Computing, 25(12), 24494-24511. https://doi.org/10.1109/tmc.2026.3735391
Vancouver
Wang W, Kim DI, Xie R, Tang Q, Xie G, Huang T. Average Response Time Optimization for Service-Chain Routing in Computing Power Networks via Robust Queueing Network Analysis. IEEE Transactions on Mobile Computing. 2026;25(12):24494-24511. doi:10.1109/tmc.2026.3735391
BibTeX
@article{Wang2026,
author = {Wang, Wenzheng and Kim, Dong In and Xie, Renchao and Tang, Qinqin and Xie, Gaochang and Huang, Tao},
title = {Average Response Time Optimization for Service-Chain Routing in Computing Power Networks via Robust Queueing Network Analysis},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {24494--24511},
doi = {10.1109/tmc.2026.3735391}
}
EUVIMED