The accelerating convergence of artificial intelligence and edge computing has sparked a recent wave of interest in edge intelligence. While pilot efforts focused on edge DNN inference serving for a single user or DNN application, scaling edge DNN inference serving to multiple users and applications is however nontrivial. In this paper, we propose an online optimization framework EdgeAdaptor for multi-user and multi-application edge DNN inference serving at scale, which aims to navigate the three-way trade-off between inference accuracy, latency, and resource cost via jointly optimizing the application configuration adaption, DNN model selection and edge resource provisioning on-the-fly. The underlying long-term optimization problem is difficult since it is NP-hard and involves future uncertain information. To address these dual challenges, we fuse the power of online optimization and approximate optimization into a joint optimization framework, via i) decomposing the long-term problem into a series of single-shot fractional problems with a regularization technique, and ii) rounding the fractional solution to a near-optimal integral solution with a randomized dependent scheme. Rigorous theoretical analysis derives a parameterized competition ratio of our online algorithms, and extensive trace-driven simulations verify that its empirical value is no larger than 1.4 in typical scenarios.

This paper studies the cloud market for computing jobs with completion deadlines, and designs efficient online auctions for cloud resource provisioning. A cloud user bids for future cloud resources to execute its job. Each bid includes: 1) a utility, reflecting the amount that the user is willing to pay for executing its job and 2) a soft deadline, specifying the preferred finish time of the job, as well as a penalty function that characterizes the cost of violating the deadline. We target cloud job auctions that executes in an online fashion, runs in polynomial time, provides truthfulness guarantee, and achieves optimal social welfare for the cloud ecosystem. Towards these goals, we leverage the following classic and new auction design techniques. First, we adapt the posted pricing auction framework for eliciting truthful online bids. Second, we address the challenge posed by soft deadline constraints through a new technique of compact exponential-size LPs coupled with dual separation oracles. Third, we develop efficient social welfare approximation algorithms using the classic primal-dual framework based on both LP duals and Fenchel duals. Empirical studies driven by real-world traces verify the efficacy of our online auction design.

Federated Learning (FL) is a new distributed machine learning (ML) approach which enables thousands of mobile devices to collaboratively train artificial intelligence (AI) models using local data without compromising user privacy. Although FL represents a promising computing paradigm, such training process can not be fully realized without an appropriate economic mechanism that incentivizes the participation of heterogeneous clients. This work targets social cost minimization, and studies the incentive mechanism design in FL through a procurement auction. Different from existing literature, we consider a practical scenario of FL where clients are selected and scheduled at different global iterations to guarantee the completion of the FL job, and capture the distinct feature of FL that the number of global iterations is determined by the local accuracy of all participants to balance between computation and communication. Our auction framework $A_{FL}$ first decomposes the social cost minimization problem into a series of winner determination problems (WDPs) based on the number of global iterations. To solve each WDP, $A_{FL}$ invokes a greedy algorithm to determine the winners, and a payment algorithm for computing remuneration to winners. Finally, $A_{FL}$ returns the best solution among all WDPs. We carried out theoretical analysis to prove that $A_{FL}$ is truthful, individual rational, computationally efficient, and achieves a near-optimal social cost. We further extend our model to consider multiple FL jobs with corresponding budgets and propose another efficient algorithm $A_{FL-M}$ to solve the extended problem. We conduct large-scale simulations based on the real-world data and testbed experiments by adopting FL frameworks FAVOR and CoCoA. Simulation and experiment results show that both $A_{FL}$ and $A_{FL-M}$ can reduce the social cost by up to 55% compared with state-of-the-art algorithms.

Effective incentive mechanisms are invaluable in mobile crowd sensing, for stimulating participation of smartphone users. Online auction mechanisms represent a natural solution for such sensing task allocation. Departing from existing studies that focus on an isolated system round, we optimize social cost across the system lifespan, while considering location constraints and capacity constraints when assigning sensing tasks to users. The winner determination problem (WDP) at each round is NP-hard even without inter-round coupling imposed by user capacity constraints. We first propose a truthful one-round auction, comprising of an approximation algorithm for solving the one-round WDP and a payment scheme for computing remuneration to winners. We then propose an online algorithm framework that employs the one-round auction as a building block towards a flexible mechanism that makes on-spot decisions upon dynamically arriving bids. Through both theoretical analysis and trace-driven simulations, we demonstrate that our online auction is truthful, individually rational, computationally efficient, and achieves a good competitive ratio.

Compared with traditional virtual machines, cloud containers are more flexible and lightweight, emerging as the new norm of cloud resource provisioning. We exploit this new algorithm design space, and propose scheduling frameworks for cloud container services. Our offline and online schedulers permit partial execution, and allow a job to specify its job deadline, desired cloud containers, and inter-container dependence relations. We leverage the following classic and new techniques in our scheduling algorithm design. First, we apply the compact-exponential technique to express and handle nonconventional scheduling constraints. Second, we adopt the primal-dual framework that determines the primal solution based on its dual constraints in both the offline and online algorithms. The offline scheduling algorithm includes a new separation oracle to separate violated dual constraints, and works in concert with the randomized rounding technique to provide a near-optimal solution. The online scheduling algorithm leverages the online primal-dual framework with a learning-based scheme for obtaining dual solutions. Both theoretical analysis and trace-driven simulations validate that our scheduling frameworks are computationally efficient and achieve close-to-optimal aggregate job valuation.

Unmanned aerial vehicles (UAVs) have emerged as a promising technology to provide low-latency mobile edge computing (MEC) services. To fully utilize the potential of UAV-assisted MEC in practice, both technical and economic challenges need to be addressed: how to optimize UAV trajectory for online task offloading and incentivize the participation of UAVs without compromising the privacy of user equipment (UE). In this work, we consider unique features of UAVs, i.e., high mobility as well as limited energy and computing capacity, and propose privacy-preserving auction frameworks, Ptero, to schedule offloading tasks on the fly and incentivize UAVs’ participation. Specifically, Ptero first decomposes the online task offloading problem into a series of one-round problems by scaling the UAV’s energy constraint into the objective. To protect UE’s privacy, Ptero calculates UAV’s coverage based on subset-anonymity. At each round, Ptero schedules UAVs greedily, computes remuneration for working UAVs, and processes unserved tasks in the cloud to maximize the system’s utility ( i.e., minimize social cost). Theoretical analysis proves that Ptero achieves truthfulness, individual rationality, computational efficiency, privacy-preserving and a nontrivial competitive ratio. Trace-driven evaluations further verify that Ptero can reduce the social cost by up to $116\%$ compared with four state-of-the-art algorithms.