TL;DR
This paper tackles the challenge of efficiently offloading tasks and caching services in mobile edge computing (MEC) to minimize user delay and balance server load. They propose a novel two-layer strategy, JTOSC, that dynamically caches services and intelligently offloads tasks, achieving significant improvements in maximum delay, cache hit rate, and load balance compared to existing methods.
What Problem Does This Paper Address?
Mobile edge computing (MEC) aims to bring computation closer to users, addressing the limitations of mobile devices and the latency of traditional cloud computing. However, a key challenge is the limited storage and computing resources of edge servers. This paper focuses on the collaborative task offloading and service caching problem in MEC. Specifically, it addresses the contradiction between the need for diverse services to be available at the edge for delay-sensitive applications (like autonomous driving or VR/AR) and the constrained storage of edge servers. Existing approaches often assume edge servers have all necessary services cached or don't adequately consider the dynamic nature of user demands and the need for fairness among users and balanced load across multiple edge servers. This can lead to task execution failures, increased delays, and poor user experience.
Methods
The paper introduces a two-layer collaborative MEC system model and a Joint Task Offloading and Service Caching (JTOSC) algorithm.
- Outer Layer (Service Caching): This layer uses Gibbs sampling to iteratively update service caching decisions. The idea is to simulate conditional samples by updating one variable (caching decision for a specific server) while keeping others fixed. This probabilistic approach helps converge to an optimal caching policy.
- Inner Layer (Task Scheduling): This layer is further decomposed into two sub-problems:
- Fairness-Perception Computing Resource Allocation: This sub-problem aims to minimize the maximum delay of all users by fairly allocating computing resources among them. It uses a bisection method to find the optimal resource allocation.
- Revenue-Preference Bilateral Matching Task Offloading: This sub-problem determines the optimal offloading location for tasks. It employs a preference-based bilateral matching algorithm that considers the estimated delay and a "revenue" (difference between maximum allowed delay and actual delay) to match tasks to the most suitable edge servers or the local device.
The JTOSC algorithm integrates these components to achieve a holistic solution.
Datasets
Not discussed in the paper. The authors use simulation-based evaluation.
Key Insights and Ideas
- The Interplay of Caching and Offloading is Crucial: The paper highlights that effective task offloading in MEC is not possible without intelligent service caching. Limited edge storage necessitates dynamic caching strategies that adapt to user demands.
- Collaboration is Key for Load Balancing and Reduced Delay: When edge servers collaborate, tasks that miss their local cache can be offloaded to other nearby MECs, significantly reducing reliance on the distant cloud and balancing the workload across servers.
- Fairness Matters for User Experience: Simply optimizing for overall system benefits can lead to unfairness. The proposed approach explicitly considers user fairness by minimizing the maximum delay experienced by any user.
- Gibbs Sampling for Dynamic Caching: Using Gibbs sampling for service caching allows for adaptive decisions, which is more suitable for dynamic user demands than static or popularity-based caching.
- Bilateral Matching for Optimal Offloading: The revenue-preference bilateral matching approach effectively balances task offloading decisions with resource availability and user delay constraints.
Key Results
The simulation results demonstrate that the proposed JTOSC strategy outperforms four other comparison strategies (CORA, JTORA, OSPRA, CDCCO) across several key metrics:
- Maximum Delay: JTOSC achieved the lowest maximum delay (0.2205s) among all strategies, indicating better fairness and reduced latency for the worst-case users.
- Local Service Cache Hit Ratio: JTOSC achieved the highest local service cache hit ratio (91.63%), signifying efficient service caching and reduced reliance on offloading to other servers or the cloud.
- Load Balance: Figures 6 and 7 show that JTOSC leads to a more balanced distribution of tasks among MEC servers compared to strategies that don't consider horizontal collaboration.
- Impact of Parameters: The study also shows that increasing computing capabilities of MEC servers reduces delay, but the benefit plateaus due to storage limitations. Similarly, increasing caching capacity reduces delay, but again, computing resources become the bottleneck.
Limitations and Future Work
- Assumptions: The paper assumes that all users share communication resources equally and ignores inter-cell interference.
- Future Work: The authors suggest that communication interference management will be a focus of their future research. They also propose that this study can serve as a reference for future task offloading research in MEC.
Summary updated