Publications-Periodical Articles

Article View/Open

Publication Export

Google ScholarTM

NCCU Library

Citation Infomation

Related Publications in TAIR

題名 Non-intrusive kernel-level dispatching for MQTT shared subscriptions
作者 廖峻鋒
Chiu, Yi-Hsiu;Liao, Chun-Feng
貢獻者 資訊系
關鍵詞 Workload dispatching; Message dispatching; MQTT shared subscriptions; eBPF; Edge computing
日期 2026-09
上傳時間 3-Aug-2026 13:24:12 (UTC+8)
摘要 Edge computing environments consist of distributed, resource-constrained networks and edge servers that collaboratively process time-sensitive data streams from sensors. To support communication in such settings, MQTT has become a widely adopted protocol due to its lightweight design and flexible topic-based publish/subscribe model. MQTT 5.0 further extends this model by introducing shared subscriptions, which allow multiple subscribers to jointly consume messages and enable implicit load balancing. However, existing MQTT brokers rely on static workload dispatching strategies, which are inadequate for dynamic environments characterized by heterogeneous resources and highly variable workloads. To address these limitations, this paper proposes NIKD (Non-Intrusive Kernel-level Dispatching), an eBPF-based framework that decouples workload dispatching logic from the broker and enables dynamic, real-time dispatching for MQTT shared subscriptions. We present NIKD_LOR, a proof-of-concept prototype that implements the Least Outstanding Request (LOR) strategy to route messages based on the number of pending requests at each subscriber. Experimental results demonstrate that, compared to native broker dispatching strategies based on static algorithms, NIKD_LOR improves workload fairness and system responsiveness, reducing average task completion time by over 20% while incurring acceptable overhead.
關聯 Future Generation Computer Systems, Vol.182, 108454
資料類型 article
DOI https://doi.org/10.1016/j.future.2026.108454
dc.contributor 資訊系
dc.creator (作者) 廖峻鋒
dc.creator (作者) Chiu, Yi-Hsiu;Liao, Chun-Feng
dc.date (日期) 2026-09
dc.date.accessioned 3-Aug-2026 13:24:12 (UTC+8)-
dc.date.available 3-Aug-2026 13:24:12 (UTC+8)-
dc.date.issued (上傳時間) 3-Aug-2026 13:24:12 (UTC+8)-
dc.identifier.uri (URI) https://ah.lib.nccu.edu.tw/item?item_id=183570-
dc.description.abstract (摘要) Edge computing environments consist of distributed, resource-constrained networks and edge servers that collaboratively process time-sensitive data streams from sensors. To support communication in such settings, MQTT has become a widely adopted protocol due to its lightweight design and flexible topic-based publish/subscribe model. MQTT 5.0 further extends this model by introducing shared subscriptions, which allow multiple subscribers to jointly consume messages and enable implicit load balancing. However, existing MQTT brokers rely on static workload dispatching strategies, which are inadequate for dynamic environments characterized by heterogeneous resources and highly variable workloads. To address these limitations, this paper proposes NIKD (Non-Intrusive Kernel-level Dispatching), an eBPF-based framework that decouples workload dispatching logic from the broker and enables dynamic, real-time dispatching for MQTT shared subscriptions. We present NIKD_LOR, a proof-of-concept prototype that implements the Least Outstanding Request (LOR) strategy to route messages based on the number of pending requests at each subscriber. Experimental results demonstrate that, compared to native broker dispatching strategies based on static algorithms, NIKD_LOR improves workload fairness and system responsiveness, reducing average task completion time by over 20% while incurring acceptable overhead.
dc.format.extent 108 bytes-
dc.format.mimetype text/html-
dc.relation (關聯) Future Generation Computer Systems, Vol.182, 108454
dc.subject (關鍵詞) Workload dispatching; Message dispatching; MQTT shared subscriptions; eBPF; Edge computing
dc.title (題名) Non-intrusive kernel-level dispatching for MQTT shared subscriptions
dc.type (資料類型) article
dc.identifier.doi (DOI) 10.1016/j.future.2026.108454
dc.doi.uri (DOI) https://doi.org/10.1016/j.future.2026.108454