We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.NI

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Networking and Internet Architecture

Title: Energy-based Proportional Fairness in Cooperative Edge Computing

Abstract: By executing offloaded tasks from mobile users, edge computing augments mobile user equipments (UEs) with computing/communications resources from edge nodes (ENs), enabling new services (e.g., real-time gaming). However, despite being more resourceful than UEs, allocating ENs' resources to a given favorable set of users (e.g., closer to ENs) may block other UEs from their services. This is often the case for most existing approaches that only aim to maximize the network social welfare or minimize the total energy consumption but do not consider the computing/battery status of each UE. This work develops an energy-based proportional-fair framework to serve all users with multiple tasks while considering both their service requirements and energy/battery levels in a multi-layer edge network. The resulting problem for offloading tasks and allocating resources toward the tasks is a Mixed-Integer Nonlinear Programming, which is NP-hard. To tackle it, we leverage the fact that the relaxed problem is convex and propose a distributed algorithm, namely the dynamic branch-and-bound Benders decomposition (DBBD). DBBD decomposes the original problem into a master problem (MP) for the offloading decisions and multiple subproblems (SPs) for resource allocation. To quickly eliminate inefficient offloading solutions, MP is integrated with powerful Benders cuts exploiting the ENs' resource constraints. We then develop a dynamic branch-and-bound algorithm (DBB) to efficiently solve MP considering the load balance among ENs. SPs can either be solved for their closed-form solutions or be solved in parallel at ENs, thus reducing the complexity. The numerical results show that DBBD returns the optimal solution in maximizing the proportional fairness among UEs. DBBD has higher fairness indexes, i.e., Jain's index and min-max ratio, in comparison with the existing ones that minimize the total consumed energy.
Subjects: Networking and Internet Architecture (cs.NI)
Cite as: arXiv:2201.00910 [cs.NI]
  (or arXiv:2201.00910v1 [cs.NI] for this version)

Submission history

From: Thai T. Vu [view email]
[v1] Mon, 3 Jan 2022 23:38:35 GMT (973kb)
[v2] Fri, 11 Feb 2022 10:52:51 GMT (973kb)

Link back to: arXiv, form interface, contact.