Search

Edge computing pushes computation and storage resources to the network edge which is close to end users, and thus is critical for latency-sensitive applications, e.g., intelligent Vehicular Ad Hoc Networks (VANETs). To enable these services, a set of edge servers need to be deployed to the roadsides. Such deployment should offer low-latency services to end users, while keeping a low deployment or maintenance cost, which is a non-trivial task. In this paper, we study the edge server placement problem in a metropolitan area. This problem is composed of two parts: to determine 1) the locations of the servers and 2) the coverage of each server, with multiple optimization objectives. First, we study the Shanghai Taxi Trace to gain insights into the traffic pattern of taxis, especially how vehicles move between different locations. Second, we build multi-objective optimization models to characterize the trade-off among three critical performance metrics, namely the initial deployment cost, the runtime cost (i.e., number of hand-offs between different servers), and the average delay of tasks. Due to the intractability of these NP-hard problems, we propose a heuristic multi-objective optimization method to decompose the global problem into a set of local problems with tractable scale. Numerical results verify that our heuristic strategy achieves a desirable balance among the three performance metrics, e.g., a 5% compromise of the delay can reduce up to 50% of the hand-offs for small local areas, and 10%+ for the entire global area, compared with the best existing algorithms.

Datacenter networks (DCNs) play a key role in providing cloud services. The energy consumption and cost of a DCN are growing sharply with the extensions of network bandwidth and network size. The energy consumption, complexity and cost of a DCN depend on some design factors such as the topology structure, addressing scheme and routing mechanism. A novel addressing and routing architecture for cloud-service DCNs with regular topologies is proposed in this paper. First of all, we propose a port-based source-routing addressing (PSRA) scheme, which makes the table-lookup operation unnecessary and decreases the switch complexity. Next, leveraging the characteristics of PSRA and the regularity of DCN topologies, an extremely simple routing mechanism is designed, without switch involvement, control message interaction and topology information storage. Lastly, a high-efficiency fault-tolerance mechanism is proposed for the addressing and routing architecture. The analysis, implementation and simulation results indicate that the proposed architecture not only decreases the energy consumption and thus the cost of a DCN, but also enhances the routing performance and solves the fault-tolerance problem in a very efficient way.

Low-latency data service is an increasingly critical challenge for data center applications. In the modern distributed storage systems, proper data placement helps reduce the data movement delay, which can contribute to the service latency reduction tremendously. Existing data placement solutions have often assumed the prior distribution of data requests or discovered it via trace analysis. However, data placement is a difficult online decision-making problem faced with dynamic network conditions and time-varying user request patterns. The conventional static model-based solutions are less effective to handle the dynamic system. With an overall consideration of data movement and analytical latency, we develop a reinforcement learning-based framework DataBot+, automatically learning the optimal placement policies. DataBot+ adopts neural networks, trained with a variant of Q-learning, whose input is the real-time data flow measurements and whose output is a value function estimating the near-future latency. For instantaneous decision making, DataBot+ is decoupled into two asynchronous production and training components, ensuring that the training delay will not introduce extra overheads to handle the data flows. Evaluation results driven by real-world traces demonstrate the effectiveness of our design.