P2P.Resource.Pool资源-CSDN文库

需积分: 3 108 浏览量 2008-03-18 10:56:08 上传评论收藏 231KB PDF 举报

### P2P.Resource.Pool及其在优化广域应用级组播中的应用 #### 概述 “P2P.Resource.Pool”这一概念主要探讨了如何利用P2P（Peer-to-Peer）技术来构建资源池，并进一步优化广域范围内的应用级组播服务（Application Level Multicasting, ALM）。该研究由微软亚洲研究院的研究人员完成，旨在解决两个关键问题：一是如何组织一个P2P资源池；二是如何将这种资源池应用于实际场景中，特别是如何优化广域范围内的应用级组播。 #### 组织P2P资源池在构建P2P资源池方面，研究人员提出了一种结合P2P分布式哈希表（Distributed Hash Table, DHT）与自适应监测基础设施的方法。这种方法充分利用了P2P网络自我组织的能力以及DHT的灵活性，同时通过在DHT之上构建一层自我扩展的监测机制来实现对资源池的有效管理。 1. **P2P分布式哈希表（DHT）**：DHT是一种分布式存储系统，能够有效地处理大规模数据分布和检索问题。它通过节点间的相互协作实现了数据的快速查找和存储，是构建P2P资源池的基础。 2. **自我扩展监测机制**：为了确保资源池能够根据实时需求进行自我调整，研究人员设计了一个基于系统的、可自动扩展的监测框架。这个框架可以实时监控资源池的状态，如可用资源的数量、类型等，并根据这些信息动态调整资源分配策略。 #### 应用案例：优化广域应用级组播研究者选择优化广域应用级组播作为展示P2P资源池潜力的典型案例。广域应用级组播是一种在网络中向多个接收端高效传输数据的技术，尤其是在跨地域的通信场景中尤为重要。然而，传统的组播技术面临着诸多挑战，例如网络带宽的高效利用、延迟问题等。 1. **单个ALM会话的优化**：通过利用资源池中的闲置资源，研究者展示了如何显著降低单个ALM会话的成本，包括减少延迟、提高传输效率等方面。 2. **多会话并发优化**：此外，他们还提出了一种纯粹基于市场驱动的方法来优化多个并发的ALM会话。这种方法允许不同优先级的会话根据其重要性自动获取相应的资源份额，从而实现了资源的高效分配。 #### 结论 “P2P.Resource.Pool”及其在优化广域应用级组播中的应用展示了P2P技术的巨大潜力。通过对P2P资源池的组织和优化，不仅能够有效提升广域范围内的数据传输效率，还能实现资源的合理分配，为未来的大规模分布式计算和通信提供了新的思路和技术支持。该研究为理解和探索P2P技术在复杂网络环境下的应用提供了宝贵的参考。通过以上分析可以看出，“P2P.Resource.Pool”不仅仅是关于技术架构的设计，更重要的是它提供了一种新的视角来看待资源管理和优化问题，这对于推动P2P技术在实际应用中的发展具有重要意义。

资源推荐

资源详情

资源评论

P2P Resource Pool and Its Application to Optimize

Wide-Area Application Level Multicasting

Zheng Zhang, Yu Chen, Shi-Ding Lin, Bo-Ying Lu, Shu-Ming Shi*, Xing Xie and Chun Yuan

Microsoft Research Asia

5F, Sigma building, No.49, Zhichun Road

Beijing, 100080, P.R.China

{zzhang, ychen, i-slin, t-bylu, xingx, cyuan}@microsoft.com

ssm01@mails.tsinghua.edu.cn

ABSTRACT

The concept of resource pool has a very long history. Propelled by

the need to share CPU cycles of supercomputers for high-

throughput computing jobs from the scientific community, the

vision is most recently explored by the advocates of Grid. On the

other hand, the advent of P2P researches has demonstrated the

feasibility of integrating potentially unlimited amount of less

powerful machines around the world. Organizing a P2P resource

pool thus becomes an interesting research topic.

This paper attempts to address two problems. The first is how to

organize a P2P resource pool, and our answer is to combine the

self-organizing strength of P2P DHT with an in-system, self-

scaling monitoring infrastructure that is layered on top of DHT.

The second question is the utility of the P2P resource pool for

interesting applications. And we choose to showcase its power by

optimizing wide-area application level multicasting (ALM), a

problem far more challenging and interesting than conventional

tasks such as massively parallel computation.

We show that utilizing spare resources in the pool results in

significant savings for single ALM session. Furthermore, we

adopt a purely market-driven approach to optimize multiple

concurrent sessions. As expected, sessions of higher priority are

given higher share of resources.

1. INTRODUCTION

The concept of resource pool has a very long history. Most

recently, the Grid community has propelled this vision by uniting

the local resource pools (i.e. cluster) spread over a dozen sites into

a global one. However, the applications are restricted mostly in

the area of high-throughput computing.

The advent of P2P paints a different direction. First of all, the

research community has been actively investigating applications

beyond that of number crunching. The scale and composition of

resources are of an entirely different nature: we are talking about

millions of desktop PCs spread widely apart. If a P2P resource

pool is attempted, it is not clear that the same technologies in Grid

can be adopted in a straightforward manner.

On the other hand, A P2P resource pool differs from typical P2P

applications in that there can be multiple active instances of

different applications running in the pool. Also, a peer in the pool

may be helping an application instance of which it is not

necessarily a member.

Broadly speaking, we define a P2P resource pool as a collection

of desktop-grade resources on the edge of the internet, and some

form of middleware is managing these resources such that they

can offer computing, storage and networking capabilities for

multiple instances of potentially different distributed applications.

This paper addresses two related and important questions with

respect to P2P resource pool: its architecture and utility to

schedule interesting applications. We propose to employ

structured P2P in the form of DHT (distributed hash table)

[25][30][36][42] to pool together potentially unlimited amount of

and widely distributed resources. However, DHT alone does not

automatically generate a resource pool. Thus, we combine the

self-organizing capability of P2P DHT with an in-system, self-

scaling monitoring infrastructure SOMO (Self-Organized

Metadata Overlay). SOMO builds a dynamic system status

database which is available internally to any peers. This database

is being continuously updated and thus creates an illusion of a

single, large resource pool.

Next, we demonstrate the utility of P2P resource pool by

describing how to schedule and optimize application-level

multicasting (ALM), an application that is far more challenging

than, for instance, massively parallel computations. By using peers

that are otherwise idle, our finding is that up to 30% improvement

can be made for small-to-medium group size. In the context of

scheduling multiple competing sessions, our approach is

remarkably simple: global scheduling is never attempted. Instead,

we adopt a market-driven approach and let each session competes

based on their priorities, armed with the information provided by

SOMO. Our results show that, as expected, sessions of higher

priority are given higher share of resources, resulting in better

performance.

The rest of the paper is organized as follows. We articulate the

need of a P2P resource pool in Section-2 and contrast it with

existing alternatives. The architecture of a P2P resource pool is

described in Section-3. For many interesting P2P applications,

there is a need to generate resource attributes that can not be

derived locally from a machine. For instance, network coordinates

and bottleneck bandwidth are necessary to evaluate potential

helping peers in ALM. This problem is addressed in Section-4.

Section-5 evaluates how P2P resource can help ALM and

provides experiment results. We discuss related work in Section 6

and conclude in Section 7.

2. MOTIVATION AND ALTERNATIVES

2.1 The argument for a P2P resource pool

Over the recent years, the P2P research community has

investigated many interesting P2P applications; ranging from

wide-area distributed storage [20][9][10], scientific computing

[18], application-level multicasting [6][4], distributed web cache

[35], searching [29] and collaborative spam fighting [43], to name

just a few. There is, however, a common assumption that underlies

all these proposals: all peers are active participants in one

application instance.

* This work was performed when the author was a part-time student at

Microsoft Research Asia.

P2P resource pool explores a different dimension in which 1)

there can be multiple and simultaneous instances of different

applications and they could potentially overlap on the resources

they are running; and 2) a peer maybe helping an application

instance of which it is not a member. The first point illustrates

aggregated power of all potential resources, and the second

reflects and extends the very collaborative principle upon which

the P2P premise is found.

For instance, all existing application-level multicasting (ALM)

algorithms assume that the only resources available are those in

the ALM session. In a collaborative environment, many other

stand-by resources could be included for an otherwise more

optimal solution. For example, Microsoft Research has five

branches across the globe, and has many thousands of machines

that are geographically distributed. At a given hour, however,

number of active video-conference sessions is likely to be only a

handful, and each session may have a small number of participants

(say less than 20).

The availability of a P2P resource pool offers new optimization

possibilities. As shown in Figure 1, when an otherwise idle but

suitable helping peer is identified, it can be integrated into a

topology with better performance. This is an actual output of our

algorithm used in this paper.

Figure 1. (a) An optimal plan for an ALM. (b) An even better

plan using helper nodes in the resource pool. Circles are the

original members of the session, and the square is an available

peer with a large degree.

The P2P resource pool has seen its first incarnation in PlanetLab

[24], a wide-area P2P testbed. On PlanetLab, researchers upload

experiments on to machines comprising the testbed that will be

run concurrently with others. Up-to-date, its scale is still limited

(220 nodes as of the writing of the paper), and thus a need to

organize the pool in a more scalable and self-organizing fashion is

not yet profound. This is one of the problem this paper attempts to

address.

2.2 Resource pool and its alternatives

We wish to give a more concrete definition of resource pool by

contrasting it against another interesting alternative: the job pool.

This is necessary because, from a high-level perspective, both are

venues to deliver the matchmaking between job and resource, and

that neither is perfect: depending on the application scenario, each

has its unique strength and weakness.

Informally speaking, a job pool is a collection of jobs and is where

an idle resource look for suitable work to perform, whereas a

resource pool is the precisely the opposite: a task manager goes

into a resource pool to discover and acquire necessary helping

hands in order to accomplish a given mission. Of course, in a

distributed environment, there can be legitimate combination of

both.

A perfect example of a p2p job pool is SETI@home [18] (and of

course, many others of the same flavor). There is a well-

maintained central site and, typically, the application should be

easily parceled out for distribution. Machines register themselves

in order to grab a piece of work and then go away cranking away

whenever they feel like. This is a very economical model and

requires very limited amount of management at the central site.

Provided that job is of coarse granularity, a centralized

architecture works extremely well. It has been reported that

SETI@home has aggregated computing power far exceeding some

of the most powerful supercomputers in the world. The limitation

is also obvious. Although it is possible to think of advanced

variations, because the unpredictability of when and what

resources will become available, applications are restricted mostly

to those that are conventionally known as “embarrassingly

parallel” ones. It is also possible to implement the job pool as a

distributed architecture, but it will be far easier to just use a

centralized architecture.

In contrast, a node joins a resource pool in the hope that its power,

when otherwise idle, can be of some use. The economic incentive

can be stronger, especially in the context of P2P: tasks of arbitrary

type (beyond those of number crunching) will tap into the power

of other participants in the pool at some suitable point. The added

flexibility is particularly useful for applications such as running

application-level multicasting sessions with some level of QoS

guarantee. This is so because planning the topology of the tree is

itself a complex piece of work. On the other hand, the

consequences are many. Foremost of all is an accurate accounting

of what is going on in the resource pool. This is necessary for

each job to quickly query the available candidates and

subsequently make resource reservation. The implication is that,

given the potentially huge amount of resources in the millions, a

client-server architecture where each client updates one central

entity about its status is no longer a scalable – not to say robust

alternative.

To summarize, job pool is best for scenarios where the task can be

well-partitioned. Resource pool can ideally accommodate tasks of

arbitrary type. However, it will need, as a minimum, a scalable

way of monitoring and aggregating system information so that

resource reservations can be carried out at the discretion of task

managers that is responsible for individual incoming jobs.

The principle of resource pool is what motivates the work in the

Grid space, in particular the Condor-G line of work [14][8]. For

instance, the Grip Resource Registration Protocol (GRRP) is used

for an entity, typically representing a cluster of machines, to notify

other entities that it is part of the pool. Grip Resource Information

Protocol (GRIP), on the other hand, is the primitive to construct

aggregated resource directory service through which tasks can

query for potential candidates. The Condor-G agent can use such

infrastructure to submit jobs and monitor its progress.

There have been many discussions about the convergence of P2P

and Grid [13][22]. We believe that indeed there are many

synergies among the two in the space of resource pool

organization. In particular, we argue that the self-organizing

attributes is what the many excellent work of P2P can bring to the

scene of Grid. We will offer a more elaborate discussion at the

conclusion of Section-3.

3. BUILDING P2P RESOURCE POOL

The foundation of our resource pool proposal is the so-called

structured peer-to-peer systems, and in particular the distributed

hash table (DHT). DHT offers a way to pool together potentially

unlimited amount of resources together. But the capacity to pool

(a)

(b)

剩余8页未读，继续阅读

评论收藏

内容反馈

ocaction

粉丝: 0
资源: 1

P2P.Resource.Pool

Erlang Resource Pool:资源池管理多个进程之间的可重用资源。-开源

java p2p.part4.rar

p2p.rar_P2P_P2P 文件传输_P2P文件传输_易语言p2p传输

P2P之UDP穿透NAT P2P.WellKnown

安卓P2P相关-Android-Sip2Peer-1.0实现p2p.zip

java p2p.part2.rar

java p2p.part1.rar

java p2p.part3.rar

p2p.zip_P2P_p2p文件分发系统_文件传输

p2p.dll.mui

P2P.WellKnown

旅行商p2p.rar

P2P.tar.gz

p2p.zip_P2P

P2P.zip_p2p VC

p2p.rar_P2P.ini_p2pclient

p2p.rar_P2P_p2p server

P2p.rar_源码

基于java的第三代的P2P网络 ANts P2P.zip

Linux运维-4-文件分发之P2P.mp4.mp4

TCP-P2P.rar_P2P nat实现C++_TCP 打洞 _p2p TCP_tcp_打洞

AppScan_Std_9.0.3.5_Eval_Win.rar.P2P.DOWNLOAD

股权众筹-P2P.rp

积木盒子：从数据入手P2P.pdf

P2P.rar_vc p2p

P2P.rar_p2p CSharp_wpf p2p

安卓开发-Android-Sip2Peer-1.0 实现p2p.zip.zip

Android-Sip2Peer-1.0 实现p2p.zip项目安卓应用源码下载

p2p.rar_visual c

P2P.rar_UDP P2P_UDP NAT_UDP P2P_vc p2p

最新资源