Online Security-Aware and Reliability-Guaranteed AI Service Chains Provisioning in Edge Intelligence Cloud
Yu Qiu, Junbin Liang, Victor C. M. Leung, Min Chen
Abstract
With the rapid development of edge intelligence cloud (EIC), mobile users are not satisfied with a single artificial intelligence inference service, but require multiple inference services with chain dependencies to process data. Each AI service chain (AISC) is provided as a series of interconnected virtual network functions (VNFs) on-demand deployed on edge servers. However, AISCs experience unpredictable failures and potential attacks in EIC, which may violate different inference requirements of mobile users for reliability, security, and accuracy. How to optimally deploy VNFs and BVNFs on trusted edge servers, and select secure links to form satisfactory AISCs, meanwhile throughput of receiving requests is maximized while deployment cost of computing resources used to create VNFs and BVNFs with different model sizes is minimized in real-time, is a challenging problem. In this paper, the problem is first formulated as an integer linear programming and proved to be NP-hard. Then, we consider the problem under two online backup scenarios: one is an on-site scenario where AISC requests from the mobile devices arrive one by one, and link securities between VNFs and corresponding BVNFs are ignored because they are always on the same edge server; another is an off-site scenario where a set of AISC requests are given, and VNFs and BVNFs are deployed on different servers. Finally, two online algorithms with provable competitive ratios are proposed to solve the above two problems in polynomial time. Theoretical analyses and experiments based on real network topologies demonstrate that our algorithms are promising compared to baseline algorithms.