Performance Of High-performance Computing Applications Research Articles

Framework for Performance Enhancement of MPI based Application on Cloud

Cloud technology is a major revolution that has happened in the computing era that has changed the way applications and resources are used. Elasticity is the key characteristic of the cloud, wherein the required number of resources are provided as a service with a pay-as-you-go principle. This reduces the huge cost involved in buying, installing, and maintaining the resources. Cloud computing, with its highly scalable resources, can be a good platform for High-Performance Computing (HPC). HPC application performance highly depends on the quantity of the resources, which makes the cloud a suitable candidate. But the HPC community is not very happy with cloud technology, and most of the users still think cloud technology is not suitable for HPC applications. But virtualization technology, which is the foundation of the cloud, degrades the performance of applications in the urge to improve utilization. The hypervisor layer and resource sharing by the virtual machines (VMs) hosted on the same node are the main reasons for performance degradation in the cloud. The majority of the HPC applications belong to the message passing (MPI) category, and for these applications, communication cost is the major stakeholder in deciding performance. If these applications are hosted on the cloud, it leads to further performance degradation as the process on a VM communicates through the virtual network interface, which in turn shares the network interface of the host machine with other VMs. MPI-based applications hosted on MPPs work in a bandwidth shared environment as multiple processes communicate over the same network. But in the cloud, as the number of VMs increases, per node bandwidth availability per communication reduces. To address the above issues, we have built a framework to enhance the performance of MPI-based HPC applications on VMs by considering proper VM placement strategy and resource reservation policies, with knowledge of resource availability and process communication patterns. A VM placement strategy for dynamic clustering of VMs with high priority for shared memory-based communication is proposed and tested. Results show that with a medium number of processes, there is an improvement of around 70% with our placement strategy for high data communicating processes. If there are fewer processes, and the single physical node can hold all the VMs, then the performance improvement is up to 500%.

Dynamic Resource Allocation Method Based on Symbiotic Organism Search Algorithm for HPC Application

Cloud computing (CC) merits such as scalability, reliability, and resource pooling have enticed experts to implement their high performance computing (HPC) applications on cloud. Nonetheless, HPC applications can pose critical defies on cloud that could emasculate the acquired merits if appropriate care is not taken. The main objective of this research is to develop a task scheduling algorithm that convalesces HPC application performance on software-defined network (SDN) with abridged cost. Moreover, a novel framework is adapted to improve HPC task scheduling performance on cloud in an SDN-enabled network, which is a fundamental part of CC and used for resource delivery. A proposed algorithm for task scheduling architecture manages scheduling tasks. Also, the proposed algorithm monitors CPU utilization, cost, and provides best optimal results when compared with conventional algorithms such as first come first served, round robin, and optimized time synchronization approach.

Optimizing I/O Performance of HPC Applications with Autotuning

Parallel Input output is an essential component of modern high-performance computing (HPC). Obtaining good I/O performance for a broad range of applications on diverse HPC platforms is a major challenge, in part, because of complex inter dependencies between I/O middleware and hardware. The parallel file system and I/O middleware layers all offer optimization parameters that can, in theory, result in better I/O performance. Unfortunately, the right combination of parameters is highly dependent on the application, HPC platform, problem size, and concurrency. Scientific application developers do not have the time or expertise to take on the substantial burden of identifying good parameters for each problem configuration. They resort to using system defaults, a choice that frequently results in poor I/O performance. We expect this problem to be compounded on exascale-class machines, which will likely have a deeper software stack with hierarchically arranged hardware resources. We present as a solution to this problem an autotuning system for optimizing I/O performance, I/O performance modeling, I/O tuning, and I/O patterns. We demonstrate the value of this framework across several HPC platforms and applications at scale.

Strategies of data layout and cache writing for input-output optimization in high performance scientific computing: Applications to the forward electrocardiographic problem.

Input-output (I/O) optimization at the low-level design of data layout on disk drastically impacts the efficiency of high performance computing (HPC) applications. However, such a low-level optimization is in general challenging, especially when using popular scientific file formats designed with an emphasis on portability and flexibility. To reconcile these two aspects, we present a novel low-level data layout for HPC applications, fully independent of the number of dimensions in the dataset. The new data layout improves reading and writing efficiency in large HPC applications using many processors, and in particular during parallel post-processing. Furthermore, its combination with a cached write mode, in order to aggregate multiple writes into larger ones, substantially decreased the writing times of the proposed strategy. When applied to our simulation framework for the forward calculation of the human electrocardiogram, the combined strategy resulted in drastic improvements in I/O performance, of up to 40% in writing and 93–98% in reading for post-processing tasks. Given the generality of the proposed strategies and scientific file formats used, our results may represent significant improvements in I/O performance of HPC applications across multiple disciplines, reducing execution and post-processing times and leading to a more efficient use of HPC resource envelopes.

Impact of Process Allocation Strategies in High Performance Cloud Computing on Azure Platform

Nowadays, there is an increasing demand in the High-Performance Computing (HPC) community to make use of different public cloud service provider. The question of which cloud provider is superior for a certain application and usage configuration is very important for the successful deployment of HPC application on the cloud. In this paper, we evaluate the performance of HPC applications on Microsoft Azure cloud platform using the well-known NAS parallel benchmarks. These benchmarks are considered as examples of general scientific HPC applications to test the communication performance. Different process allocation strategies are performed in terms of MOPS and Speedup. Our results show that allocating one process per instance achieves higher scalability at the expense of the cost. The results compared with the same results with the same experiments in Amazon platform. We found that Azure platform has better shared-memory communication performance than Amazon platform. In contrast, Amazon is superior to Azure platform in terms of Ethernet bandwidth.

Nosv: A lightweight nested-virtualization VMM for hosting high performance computing on cloud

Abstract Moving the high performance computing (HPC) to Cloud not only reduces the costs but also gives users the ability to customize their system. Besides, compared with the traditional HPC computing environments, such as grid and cluster which run HPC applications on bare-metal, cloud equipped with virtualization not only improves resource utilization but also reduces maintenance cost. While for some reasons, current virtualization-based cloud has limited performance for HPC. Such performance overhead could be caused by Virtual Machine Monitor (VMM) interceptions, virtualized I/O devices or cross Virtual Machine (VM) interference, etc. In order to guarantee the performance of HPC applications on Cloud, the VMM should interfere guest VMs as less as possible and allocate dedicated resources such as CPU cores, DRAM and devices to guest VMs running HPC applications. In this paper, we propose a novel cloud infrastructure to serve the HPC applications and normal applications concurrently. This novel infrastructure is based on a lightweight high performance VMM named nOSV. For HPC applications, nOSV constructs a strong isolated high performance guest VM with dedicated resources. At runtime, the high performance VM manages all resources by itself and is not interfered by nOSV. By supporting nested virtualization, nOSV can run HPC with commodity application and keep the flexibility of traditional Cloud. nOSV runs other virtualization environments, like Xen and Docker, as high performance guest VMs. All commodity Cloud applications are hosted in these virtualization environments and share hardware resources with each other. The prototype of nOSV shows that it provides a bare-metal like performance for HPC applications and has about 23% improvement compared to HPC applications running on Xen.

Evaluating and Improving the Performance and Scheduling of HPC Applications in Cloud

Cloud computing is emerging as a promising alternative to supercomputers for some high-performance computing (HPC) applications. With cloud as an additional deployment option, HPC users and providers are faced with the challenges of dealing with highly heterogeneous resources, where the variability spans across a wide range of processor configurations, interconnects, virtualization environments, and pricing models. In this paper, we take a holistic viewpoint to answer the question— why and who should choose cloud for HPC, for what applications, and how should cloud be used for HPC? To this end, we perform comprehensive performance and cost evaluation and analysis of running a set of HPC applications on a range of platforms, varying from supercomputers to clouds. Further, we improve performance of HPC applications in cloud by optimizing HPC applications’ characteristics for cloud and cloud virtualization mechanisms for HPC. Finally, we present novel heuristics for online application-aware job scheduling in multi-platform environments. Experimental results and simulations using CloudSim show that current clouds cannot substitute supercomputers but can effectively complement them. Significant improvement in average turnaround time (up to 2X)and throughput (up to 6X) can be attained using our intelligent application-aware dynamic scheduling heuristics compared tosingle-platform or application-agnostic scheduling.

Improving HPC Application Performance in Public Cloud

Improving as well as evaluating the performance of High Performance Computing (HPC) applications by migrating them to Cloud environments are widely considered as critical issues in the field of high performance and Cloud computing. However, poor network performance, heterogeneous and dynamic environments are some series of pitfalls for execution of HPC applications in Cloud. This paper proposes a new approach to improve the performance and scalability of HPC applications on Amazon's HPC Cloud. The evidence from our approach points a significant improvement in speed up and scale up with the response rate of more than 20 percent parallel efficiency on the Cloud in comparison to dedicated HPC cluster. We state that the EC2 Cloud system is a feasible platform for deploying on-demand, small sized HPC applications.

A new proposal to deal with congestion in InfiniBand-based fat-trees

The overall performance of High-Performance Computing applications may depend largely on the performance achieved by the network interconnecting the end-nodes; thus high-speed interconnect technologies like InfiniBand are used to provide high throughput and low latency. Nevertheless, network performance may be degraded due to congestion; thus using techniques to deal with the problems derived from congestion has become practically mandatory. In this paper we propose a straightforward congestion-management method suitable for fat-tree topologies built from InfiniBand components. Our proposal is based on a traffic-flow-to-service-level mapping that prevents, as much as possible with the resources available in current InfiniBand components (basically Virtual Lanes), the negative impact of the two most common problems derived from congestion: head-of-line blocking and buffer-hogging. We also provide a mathematical approach to analyze the efficiency of our proposal and several ones, by means of a set of analytical metrics. In certain traffic scenarios, we observe up to a 68% of the ideal performance gain that could be achieved in HoL-blocking and buffer-hogging prevention.

멀티-코어 서버의 성능 분석 및 특성화

멀티-코어 프로세서는 최근 마이크로프로세서 시장의 주류 제품으로 자리 잡았다. 이러한 멀티-코어 프로세서를 기반으로 하는 서버들은 고성능 컴퓨팅 분야와 상용 응용 프로그램 분야에서 그 사용 범위를 넓혀가고 있다. 멀티-코어 프로세서는 높아진 병렬성으로 인하여 응용 프로그램의 성능도 한 단계 더 높여줄 것으로 기대된다. 하지만, 칩 내부의 여러 코어들이 공유 자원들을 사용하면서 발생하는 경쟁과 충돌이 성능에 병목으로 작용하기도 한다. 그러므로 멀티-코어 서버 상에서 높은 성능과 확장성을 얻기 위해서는 공유 자원의 사용을 최적화 하는 것이 필수적이다. 본 논문에서는 코어들 간의 공유 자원 사용에서 발생하는 긍정적/부정적인 효과들이 실제 응용 프로그램의 성능에 어떻게 반영되는지 실험을 통하여 분석해 본다. 또한 이러한 분석을 통하여 멀티-코어 서버의 성능을 특성화한다. Multi-Core processors have become main-stream microprocessors in recent years. Servers based on these multi-core processors are widely adopted in High Performance Computing (HPC) and commercial business applications as well. These servers provide increased level of parallelism, thus can potentially boost the performance for applications. However, the shared resources among multiple cores on the same chip can become hot spots and act as performance bottlenecks. Therefore it is essential to optimize the use of shared resources for high performance and scalability for the multi-core servers. In this paper, we conduct experimental studies to analyze the positive and negative effects of the resource sharing on the performance of HPC applications. Through the analyses we also characterize the performance of multi-core servers.