I/O加速技术在集群中的性能优势分析

152 浏览量更新于2024-08-25 收藏 271KB PDF 举报

"I/O Acceleration Technology在集群中的优势——计算机科学" I/O Acceleration Technology (I/OAT)是由Intel开发的一系列特性，旨在显著降低网络数据处理中的接收端包处理开销。在多吉比特数据速率下，TCP/IP协议栈中的包处理占据了系统资源的相当一部分，尽管已经有许多技术用于减轻发送端的包处理负担，但接收端仍然是一个性能瓶颈。I/OAT的引入就是为了改变这一状况。 I/OAT的核心目标是通过优化硬件和软件接口来提高系统效率，特别是在高带宽环境下的网络通信。它通过减少CPU对网络数据的直接处理，将部分工作负载转移到专用的硬件加速器上，从而释放CPU资源，提高整体性能。这不仅有助于提升应用程序的运行速度，还有助于增加网络带宽利用率。本论文通过深入的微基准测试和两个不同应用领域的评估（1）多层数据中心环境和（2）并行虚拟文件系统（PVFS），研究了I/OAT技术的益处。微基准测试表明，与传统通信方式相比，I/OAT能够使总体CPU利用率降低38%。这意味着，由于CPU利用率的降低，I/OAT能提供更好的性能表现，并且可以提高网络吞吐量。在多层数据中心环境中，I/OAT有助于优化服务器之间的交互，尤其是在处理大量并发请求时，可以显著减少延迟，提高服务响应速度。而在并行虚拟文件系统（PVFS）的应用场景下，I/OAT能够加速大规模数据的读写操作，这对于依赖高速数据存取的高性能计算和大数据分析任务尤其关键。此外，I/OAT还可能带来节能效果，因为减少了CPU的工作负荷，意味着降低了能耗。对于需要持续运行的大型数据中心而言，这不仅可以节省运营成本，还能对环境产生积极影响。 I/OAT技术在集群环境中提供了显著的性能提升，通过减轻CPU负担，优化网络通信，以及增强高带宽应用的处理能力，它对于提升整个系统的效率和可扩展性具有重要意义。在当前大数据和云计算时代，这种技术的使用将对提升服务质量、降低运营成本和实现更高效的计算资源利用产生深远影响。

Beneﬁts of I/O Acceleration Technology (I/OAT) in Clusters

∗

Karthikeyan Vaidyanathan Dhabaleswar K. Panda

Computer Science and Engineering,

The Ohio State University

{vaidyana, panda}@cse.ohio-state.edu

Abstract

Packet processing in the TCP/IP stack at multi-Gigabit data rates

occupies a signiﬁcant portion of the system overhead. Though there

are several techniques to reduce the packet processing overhead on

the sender-side, the receiver-side continues to remain as a bottleneck.

I/O Acceleration Technology (I/OAT), developed by Intel, is a set of

features particularly designed to reduce the receiver-side packet pro-

cessing overhead. This paper studies the beneﬁts of the I/OAT tech-

nology by extensive evaluations through micro-benchmarks as well

as evaluations on two different application domains: (1) A multi-

tier data-center environment and (2) A Parallel Virtual File System

(PVFS). Our micro-benchmark evaluations show that I/OAT results

in 38% lower overall CPU utilization in comparison with traditional

communication. Due to this reduced CPU utilization, I/OAT delivers

better performance and increased network bandwidth. Our experi-

mental results with data-centers and ﬁle systems reveal that I/OAT

can improve the total number of transactions processed by 14% and

throughput by 12%, respectively. In addition, I/OAT can sustain a

large number of concurrent threads (up to a factor of four as com-

pared to non-I/OAT) in data-center environments, thus increasing the

scalability of the servers.

1 Introduction

Over the past few years, there has been an incredible growth

of highly data-intensive applications in various ﬁelds such as

medical informatics, genomics, e-commerce, data mining and

satellite weather image analysis. With technology trends, the

ability to store and share the datasets generated by these appli-

cations is also increasing, allowing scientists and institutions

to create large dataset repositories and making them available

for use by others. On the other hand, clusters consisting of

commodity off-the-shelf hardware components have become

increasingly attractive as platforms for high-performance com-

putation and scalable servers. Based on these two trends, re-

searchers have proposed the feasibility and potential of cluster-

based servers [14, 10, 18, 19].

Several clients request these servers for either the raw

or some kind of processed data simultaneously. How-

ever, existing servers are becoming increasingly incapable of

∗

This research is supported in part by NSF grants #CNS-0403342

and #CNS-0509452; DOE grants #DE-FC02-06ER25749 and #DE-FC02-

06ER25755; grants from Intel, Mellanox, Cisco systems, Linux Networx and

Sun Microsystems; and equipment donations from Intel, Mellanox, AMD, Ap-

ple, Appro, Dell, Microway, PathScale, IBM, SilverStorm and Sun Microsys-

tems.

meeting such sky-rocketing processing demands with high-

performance and scalability. These servers rely on TCP/IP

for data communication and typically use Gigabit Ethernet

networks for cost-effective designs. The host-based TCP/IP

protocols on such networks have high CPU utilization and

low bandwidth, thereby limiting the maximum capacity (in

terms of requests they can handle per unit time). Alternatively,

many servers use multiple Gigabit Ethernet networks to cope

with the network trafﬁc. However, at multi-Gigabit data rates,

packet processing in the TCP/IP stack occupies a signiﬁcant

portion of the system overhead.

Packet processing [12, 13] usually involves manipulating

the headers and moving the data through the TCP/IP stack.

Though this does not require signiﬁcant computation, pro-

cessor time gets wasted due to delays caused by latency of

memory accesses and data movement operations. To over-

come these overheads, researchers have proposed several tech-

niques [9] such as transport segmentation ofﬂoad (TSO),

jumbo frames, zero-copy data transfer (sendﬁle()), interrupt

coalescing, etc. Unfortunately, many of these techniques are

applicable only on the sender side, while the receiver side con-

tinues to remain as a bottleneck in several cases, thus result-

ing in a huge performance gap between the CPU overheads of

sending and receiving packets.

Intel’s I/O Acceleration Technology (I/OAT) [1, 3, 2, 15] is

a set of features which attempts to alleviate the receiver packet

processing overheads. It has three additional features, namely:

(i) split headers, (ii) DMA copy ofﬂoad engine and (iii) multi-

ple receive queues.

At this point, the following open questions arise:

• What kind of beneﬁts can be expected from the current

I/OAT architecture?

• How does this beneﬁt translate to applications?

In this paper, we focus on the above questions. We ﬁrst

analyze the performance of I/OAT based on a detailed suite

of micro-benchmarks. Next, we evaluate it on two different

application domains:

• A multi-tier Data-Center environment

• A Parallel Virtual File System (PVFS)

Our micro-benchmark evaluations show that I/OAT reduces the

overall CPU utilization signiﬁcantly, up to 38%, as compared

下载后可阅读完整内容，剩余9页未读，立即下载

weixin_38618784

粉丝: 11
资源: 884

I/O加速技术在集群中的性能优势分析

Intel IO Acceleration Technology (IOAT) Overview-计算机科学

Intel IO Acceleration Technology Overview (2006)-计算机科学

Accelerating Network Receive Processing - Intel IO Acceleration Technology (ols2005v1-pages-289-296)-计算机科学

canny检测matlab代码-FPGA-Acceleration-of-Canny-Edge-Detection-Algorithm:基于硬

Acceleration-Of-2D-Convolution-Using-Integral-Image:使用积分图像加速2D卷积

Network IO Acceleration in Heterogeneous Multicore Processors.pdf

100-GeV large scale laser plasma electron acceleration by a multi-PW laser

Magnetic field annihilation and charged particle acceleration in ultra-relativistic laser plasmas

Intel Pro-1000 PT Quad Port Bypass Server Adapter-计算机科学

The-Project-Acceleration-and-Progression-Monitoring-System-for-Software-Engineering-Project-Softwa

最新资源