Simon See

Understanding Off-Chip Memory Contention of Parallel Programs in Multicore Systems

Memory contention is an important performance issue in current multicore architectures. In this paper, we focus on understanding how off-chip memory contention affects the performance of parallel applications. Using measurements conducted on state-of-the-art multicore systems, we observed that off-chip memory traffic is not always bursty, as it was previously reported in literature. Burstiness depends on the problem size. Small problem sizes lead to bursty memory traffic, and generate small off-chip contention. In contrast, when large program sizes cause memory contention, the memory traffic is non-bursty. Based on these observations, we propose an analytical model that relates the growth of memory contention to the number of active cores and to the problem size, for both uniform (UMA) and non-uniform memory access (NUMA) systems. Our model differs from measurements on average by less than 14\%. Contention for off-chip memory grows exponentially with the number of active cores, but adding additional memory controllers reduces the memory contention. For programs such as the penta diagonal solver SP from NPB benchmark, with a large matrix of

Failure Prediction of Data Centers Using Time Series and Fault Tree Analysis

162^3

Learning Common and Specific Features for RGB-D Semantic Segmentation with Deconvolutional Networks

elements (input size C), our analysis shows that memory contention increases the total number of processor cycles to execute the program by more than ten times on a machine with 24 cores.

An improved game-theoretic approach to uncover overlapping communities

This paper proposes a framework for online failure prediction of data centers. A data center often has a high failure rate as it features a number of servers and components. Moreover, long running applications and intensive workloads are common in such facilities. Performance of the system depends on the availability of the machines, which can be easily compromised if failure cannot be handled gracefully. The main idea of this paper is to create an effective prediction model focusing on hardware failure. Accurate prediction may enhance the overall system performance. In this work, we employ two methods, namely, ARMA (Auto Regressive Moving Average) and Fault Tree Analysis. Experiments were then performed on a simulated cluster built based on Simis platform. The results show prediction accuracy of 97%, which is very high. We thus believe that our framework is practical and can be adapted to use in data centers in the future.

Galactica: A GPU Parallelized Database Accelerator

In this paper, we tackle the problem of RGB-D semantic segmentation of indoor images. We take advantage of deconvolutional networks which can predict pixel-wise class labels, and develop a new structure for deconvolution of multiple modalities. We propose a novel feature transformation network to bridge the convolutional networks and deconvolutional networks. In the feature transformation network, we correlate the two modalities by discovering common features between them, as well as characterize each modality by discovering modality specific features. With the common features, we not only closely correlate the two modalities, but also allow them to borrow features from each other to enhance the representation of shared information. With specific features, we capture the visual patterns that are only visible in one modality. The proposed network achieves competitive segmentation accuracy on NYU depth dataset V1 and V2.

A Robust Energy-Efficient Framework for Heterogeneous Datacenters

How can we uncover overlapping communities from complex networks to understand the inherent structures and functions? Chen et al. firstly proposed a community game (Game) to study this problem, and the overlapping communities have been discovered when the game is convergent. It is based on the assumption that each vertex of the underlying network is a rational game player to maximize its utility. In this paper, we investigate how similar vertices affect the formation of community game. The Adamic–Adar Index (AA Index) has been employed to define the new utility function. This novel method has been evaluated on both synthetic and real-world networks. Experimental study shows that it has significant improvement of accuracy (from 4.8% to 37.6%) compared with the Game on 10 real networks. It is more efficient on Facebook networks (FN) and Amazon co-purchasing networks than on other networks. This result implicates that “friend circles of friends” of Facebook are valuable to understand the overlapping community division.

Fast MPEG-CDVS Encoder With GPU-CPU Hybrid Computing

The amount of business data generated and collected is increasing exponentially every year. A Graphics Processing Unit (GPU) is not used for only optimization of image filtering and video processing, but is also widely adopted for accelerating big data analytics for scientific, engineering, and enterprise applications. However, there are studies pointing out that using GPU as a general-purpose computing device has limitations. In order to exploit current GPU computing capabilities for database operations, we have to take into consideration the characteristics of the GPU and how it can cooperate with the CPU. In this paper, we proposed and implemented a GPU database accelerator, which named Galactica. The experiments result shows proposed GPU database accelerator has outperformed traditional database system. In addition, the Galacticas performance is comparable with a seven nodes distributed Hadoop system. Our results indicate that the GPU is an effective and energy efficient coprocessor for executing database operations.

Real-time GPU-accelerated social media sentiment processing and visualization

Datacenters are facilities used to house computer systems. These facilities generally consume a large amount of energy. In recent years, many researches proposed datacenter management frameworks that allow energy to be utilized more efficiently. However, most of these frameworks were limited by constraints related to unpredictable behaviors of applications in both the perspectives of execution time and power consumption. In order to provide an efficient task scheduling in datacenters, this paper proposes a preliminary concept called a robust energy-efficient framework. In this framework, a software system is deployed on top of a datacenter middleware to oversee process migrations among heterogeneous machines with various configurations. Moreover, the framework integrates additional subsystems for tracking behavioral changes of scheduled processes. During runtime, these subsystems periodically generate profiles from monitored performance metrics of processes and machines. Process profiles represent resource-usage behavior of an application, while machine profiles represent resource-provisioning behaviors. Processes can be moved around on the fly based on information provided in these profiles. The proposed framework takes advantage of heterogeneity along with process migration to improve energy efficiency of a datacenter without prior knowledge on process behavior and resource usage fluctuation in users’ applications.

Artificial Intelligence Computing for a Smart City

The compact descriptors for visual search (CDVS) standard from ISO/IEC moving pictures experts group has succeeded in enabling the interoperability for efficient and effective image retrieval by standardizing the bitstream syntax of compact feature descriptors. However, the intensive computation of a CDVS encoder unfortunately hinders its widely deployment in industry for large-scale visual search. In this paper, we revisit the merits of low complexity design of CDVS core techniques and present a very fast CDVS encoder by leveraging the massive parallel execution resources of graphics processing unit (GPU). We elegantly shift the computation-intensive and parallel-friendly modules to the state-of-the-arts GPU platforms, in which the thread block allocation as well as the memory access mechanism are jointly optimized to eliminate performance loss. In addition, those operations with heavy data dependence are allocated to CPU for resolving the extra but non-necessary computation burden for GPU. Furthermore, we have demonstrated the proposed fast CDVS encoder can work well with those convolution neural network approaches which enables to leverage the advantages of GPU platforms harmoniously, and yield significant performance improvements. Comprehensive experimental results over benchmarks are evaluated, which has shown that the fast CDVS encoder using GPU-CPU hybrid computing is promising for scalable visual search.

Parameter Prediction in Fault Management Framework

Data visualization is an important aspect of data analytics in an age where decisions are all based on information. Approaches in data visualization, particularly those that have the capability of processing large-scale textual datasets and visualize them as structured information in real-time can be useful for monitoring trends in social media. In this article, we present our GPU accelerated project, which uses CUDA to distribute and parallelize the processing and analysis of textual data in order to visualize information in real-time, or close to real-time as a foundational system for the future of real-time applications which monitors trends in social media, applicable to political elections, social media analytics, and other needs in computational social sciences which are time-critical.