Search

International audience; Xen is one of the most popular virtualization platforms nowadays, which has been broadly used by the industry. Credit scheduler, the default scheduler of Xen, was initially designed for serial jobs, which achieves good performance overall for serial jobs. Unfortunately, the parallel jobs are likely to co-exist with serial jobs in the same host in practice, the resource contention between virtual machines results in severe performance degradation of the parallel jobs. In this paper, we propose vGrouper, a progressive solution to enhance the performance of the parallel jobs. The vGrouper focuses on synchronizing the execution time of the parallel nodes in order to achieve the best performance of the parallel job. Moreover, the vGrouper guarantees that the parallel job nodes are able to run concurrently on pCPUs for the entire time slice, which maximizes the efficiency of communication between parallel nodes. A prototype of vGrouper is implemented, the experimental results demonstrate that the performance of the parallel job and resource utilization in Xen have been significantly improved.

International audience; Wireless local area network (WLAN) based indoor localization is expanding its fast-paced adoption to facilitate a variety of indoor location-based services (ILBS). Unfortunately, the performance of current WLAN localization systems relying on fixed access points (APs) deployment is constrained by the spatial localizability variance (SLV) problem that different locations may exhibit significantly distinct localization resolution. Prior approaches tackle this problem through nomadic APs with favorable mobility to dynamically adjust the network topology. However, the lack of prior knowledge of nomadic AP’s position has been a challenge for location distinction and will lead to prohibitive performance degradation. In this paper, we propose and develop CNLoc, a novel CSI-based (Channel State Information) indoor WLAN localization framework to overcome the location uncertainty of nomadic APs. Our implementation and evaluation show that CNLoc can improve the accuracy with unknown location information of nomadic APs. We also discuss some open issues and new possibilities in future nomadic AP based indoor localization.

International audience; The performance issue of HDFS has always been a great concern due to its widely adoption in both production and research environments. However, a fine-grained performance analysis tool is missing to effectively identify the bottlenecks as well as to provide useful guidance for performance optimization. In this paper, we propose a fine-grained performance bottleneck analysis tool, which extends HTrace with fine-grained instrumentation points that are missing in Hadoop official distribution. In addition, we propose an effective trace merging method that improves the understandability of our analysis. We analyze the performance of HDFS under different kinds of workloads and get undiscovered insights.

International audience; In HPC systems, rich metadata are defined to describe rich information about data files, like the executions that lead to the data files, the environment variables, and the parameters of all executions, etc. Recent studies have shown the feasibility of using property graph to model rich metadata and utilizing graph traversal to query rich metadata stored in the property graph. We propose to utilize GPU to process the rich metadata graphs. There are generally two challenges to utilize GPU for metadata graph query. First, there is no proper data representation for the metadata graph on GPU yet. Second, there is no optimization techniques specifically for metadata graph traversal on GPU neither. In order to tackle these challenges, we propose GRAM, a GPU-based property graph traversal and query framework. GRAM uses GPU to express metadata graph in Compressed Sparse Row (CSR) format, and uses Structure of Arrays (SoA) layout to store properties. In addition, we propose two new optimizations, parallel filtering and basic operations merging, to accelerate the metadata graph traversal. Our evaluation results show that GRAM can be effectively applied to user scenarios in HPC systems, and the performance of metadata management is greatly improved.