Intelligent Task Caching in Edge Cloud via Bandit Learning

Area of Science:

• Edge computing
• Cloud computing
• Mobile computing

Background:

• Current task caching strategies in edge clouds often assume known user request patterns, neglecting user specificity like mobility and personalized demands.
• Existing methods overlook the impact of task size and computational requirements on caching efficiency.
• Latency is a critical factor for computation-intensive and data-intensive tasks, such as augmented reality.

Purpose of the Study:

• To address limitations in current task caching strategies by developing a more adaptive and personalized approach.
• To formalize the task caching problem and minimize task latency considering user-specific and task-specific factors.
• To introduce a novel intelligent task caching algorithm that dynamically adjusts to real-time conditions.

Main Methods:

• Formalized task caching as a non-linear integer programming problem to minimize latency.
• Developed a novel intelligent task caching algorithm named M-adaptive upper confidence bound (M-AUCB), based on multiarmed bandit principles.
• Proved that the M-AUCB algorithm achieves a sublinear regret bound, ensuring efficient learning.

Main Results:

• The M-AUCB algorithm effectively learns task patterns of mobile device requests online.
• The caching strategy dynamically incorporates task size and computational load.
• Experimental results demonstrate a significant reduction in average task latency, at least 14.8% compared to other schemes.

Conclusions:

• The M-AUCB algorithm offers a superior approach to task caching in edge cloud environments.
• By personalizing caching strategies and considering task attributes, significant latency reductions are achievable.
• This research advances edge computing by providing a more realistic and effective task caching solution.