Search

Abstract Edge-cloud applications are rapidly prevailing in recent years and pose the challenge of using both resource-strenuous edge devices and elastic cloud resources under dynamic workloads. Efficient resource allocation on edge-cloud jobs via cluster schedulers (e.g. Kubernetes/Volcano scheduler) is essential to guarantee their performance, e.g. tail latency, and such allocation is sensitive to scheduler configurations such as applied scheduling algorithms and task restart/discard policy. Deep reinforcement learning (DRL) is increasingly applied to optimize scheduling decisions. However, DRL faces the conundrum of achieving high rewards at a dauntingly long training time (e.g. hours or days), making it difficult to tune the scheduler configurations online in accordance to dynamically changing edge-cloud workloads and resources. For such an issue, this paper proposes EdgeTuner, a fast scheduler configuration tuning approach that efficiently leverages DRL to reduce tail latency of edge-cloud jobs. The enabling feature of EdgeTuner is to effectively simulate the execution of edge-cloud jobs under different scheduler configurations and thus quickly estimate these configurations’ influence on job performance. The simulation results allow EdgeTuner to timely train a DRL agent in order to properly tune scheduler configurations in dynamic edge-cloud environment. We implement EdgeTuner in both Kubernetes and Volcano schedulers and extensively evaluate it on real workloads driven by Alibaba production traces. Our results show that EdgeTuner outperforms prevailing scheduling algorithms by achieving much lower tail latency while accelerating DRL training speed by an average of 151.63x.

In recent years, Kubernetes (K8s) has become a dominant resource management and scheduling system in the cloud. In practical scenarios, short-running cloud workloads are usually scheduled through different scheduling algorithms provided by Kubernetes. For example, artificial intelligence (AI) workloads are scheduled through different Volcano scheduling algorithms, such as GANG_MRP, GANG_LRP, and GANG_BRA. One key challenge is that the selection of scheduling algorithms has considerable impacts on job performance results. However, it takes a prohibitively long time to select the optimal algorithm because applying one algorithm in one single job may take a few minutes to complete. This poses the urgent requirement of a simulator that can quickly evaluate the performance impacts of different algorithms, while also considering scheduling-related factors, such as cluster resources, job structures and scheduler configurations. In this paper, we design and implement a Kubernetes simulator called K8sSim, which incorporates typical Kubernetes and Volcano scheduling algorithms for both generic and AI workloads, and provides an accurate simulation of their scheduling process in real clusters. We use real cluster traces from Alibaba to evaluate the effectiveness of K8sSim, and the evaluation results show that (i) compared to the real cluster, K8sSim can accurately evaluate the performance of different scheduling algorithms with similar CloseRate (a novel metric we define to intuitively show the simulation accuracy), and (ii) it can also quickly obtain the scheduling results of different scheduling algorithms by accelerating the scheduling time by an average of 38.56×.

Energy consumption is a major concern in context-aware smartphone sensing. This paper first studies mobile device-based battery modeling, which adopts the kinetic battery model (KiBaM), under the scope of battery non-linearities with respect to variant loads. Second, this paper models the energy consumption behavior of accelerometers analytically and then provides extensive simulation results and a smartphone application to examine the proposed sensor model. Third, a Markov reward process is integrated to create energy consumption profiles, linking with sensory operations and their effects on battery non-linearity. Energy consumption profiles consist of different pairs of duty cycles and sampling frequencies during sensory operations. Furthermore, the total energy cost by each profile is represented by an accumulated reward in this process. Finally, three different methods are proposed on the evolution of the reward process, to present the linkage between different usage patterns on the accelerometer sensor through a smartphone application and the battery behavior. By doing this, this paper aims at achieving a fine efficiency in power consumption caused by sensory operations, while maintaining the accuracy of smartphone applications based on sensor usages. More importantly, this study intends that modeling the battery non-linearities together with investigating the effects of different usage patterns in sensory operations in terms of the power consumption and the battery discharge may lead to discovering optimal energy reduction strategies to extend the battery lifetime and help a continual improvement in context-aware mobile services.

Data center networks (DCNs) for 5G are expected to support a large number of different bandwidth-hungry applications with exploding data, such as real-time search and data analysis. As a result, significant challenges are imposed to identify the cause of link congestion between any pair of switch ports that may severely damage the overall network performance. Generally, it is expected that the granularity of the flow monitoring to diagnose network congestion in 5G DCNs needs to be down to the flow level on a physical port of a switch in real time with high-estimation accuracy, low-computational complexity, and good scalability. In this paper, motivated by a comprehensive study of a real DCN trace, we propose two sketch-based algorithms, called α-conservative update (CU) and P(d)-CU, based on the existing CU approach. α-CU adds no extra implementation cost to the traditional CU, but successfully trades off the achieved error with time complexity. P(d)-CU fully considers the amount of skew for different types of network services to aggregate traffic statistics of each type of network traffic at an individual, horizontally partitioned sketch. We also introduce a way to produce the real-time moving average of the reported results. By theoretical analysis and sufficient experimental results on a real DCN trace, we extensively evaluate the proposed and existing algorithms on their error performance, recall, space cost, and time complexity.

The Internet of Things (IoT) is a dynamic global information network consisting of Internet-connected objects, such as radio frequency identifications, sensors, and actuators, as well as other instruments and smart appliances that are becoming an integral component of the Internet. Over the last few years, we have seen a plethora of IoT solutions making their way into the industry marketplace. Context-aware communications and computing have played a critical role throughout the last few years of ubiquitous computing and are expected to play a significant role in the IoT paradigm as well. In this paper, we examine a variety of popular and innovative IoT solutions in terms of context-aware technology perspectives. More importantly, we evaluate these IoT solutions using a framework that we built around well-known context-aware computing theories. This survey is intended to serve as a guideline and a conceptual framework for context-aware product development and research in the IoT paradigm. It also provides a systematic exploration of existing IoT products in the marketplace and highlights a number of potentially significant research directions and trends.