159 163

ISSN: 2278 – 1323
International Journal of Advanced Research in Computer Engineering & Technology
Volume 1, Issue 5, July 2012

Evaluating the Performance and Delay in Unstructured
Peer to Peer Networks
K L Haritha and B Lalitha

Abstract: Peer-to-Peer computing has become a popular Consider an unstructured overlay network represented as
networking paradigm for file sharing, distributed computing, an undirected graph G= (V, E), where the set of nodes (i.e.,
collaborative working, etc. The widely used unstructured Peer to participating peers) and edges (i.e., overlay links) between
Peer protocols mainly face problems affecting their working nodes are denoted by V and E, respectively. Any node v in V
efficiency. Topology mismatch problem between the Overlay
network and its underlying network. We propose a novel
may flood messages to the nodes elsewhere in the system.
topology Self-Adaptive Topology Matching. In this method, each Given G= (V, E), we aim to improve G such that the message
joining peer is initially guided to find a physically close neighbor flooding in G becomes efficient. By efficiency, we mean to
to connect with each other. After then, its overlay location is reduce the broadcast delay from any node v to any node u V-
adaptively adjusted whenever a location mismatch is detected. {v}. Particularly, we intend to minimize the following:
The schemes to overcome the mismatch problem between the :
overlay network and the physical network, reduce the
transmission latency, and provide a fully cooperative and
reliable P2P environment. The results indicate that it usually
required shorter time to obtain better results than the other (1)
considered methods, specially for large scale problems.

Keywords: Unstructured peer-to-peer systems, topology Where P(v~u) G represents the set of all paths induced by
mismatch, Overlay Topology, broadcast..
the flooded messages due to v toward u, and lp denotes the
total delay required for traversing the path p P(v~u)
1 INTRODUCTION
Let p be any path in P (v~u) , given any v and u € V-
Peer-to-peer (P2P) model is quickly emerging as a
{u}.Denote the number of edges (or the hop count) on p by
significant computing paradigm of the future Internet. Unlike
|p|. Our second objective in this study is to minimize |p| as
traditional distributed computing, P2P networks aggregate
much as possible. Precisely, we are to minimize the following
large number of computers and possibly mobile or hand-held
devices, which join and leave the network frequently. Nodes
in a P2P network, called peers, play a variety of roles in their
interaction with other peers. When accessing information,
they are clients. When serving information to other peers, (2)
they are servers. When forwarding information for other
peers, they are routers. This new breed of systems creates
application-level virtual networks with their own overlay
topology and routing protocols. An overlay topology provides
mechanisms to create and maintain the connectivity of an
individual node to the network by establishing network
connections with a subset of other nodes (neighbors) in the
overlay network.
The P2P routing protocols allow individual computers and
devices to share information and resources directly, without
dedicated servers. Although P2P networking technologies
may provide some desirable system properties for supporting
pervasive and cooperative application sharing across the
Internet, such as anonymity, fault tolerance, low maintenance
& low administration cost, and transparent & dynamic
operability, there are some well-known problems with most of Fig.1. An example of an unstructured P2P network G=(V,E)
the current P2P systems. For the P2P paradigm to be adopted V={A,B,C,D,E,F,G,H},E={AB,AF,AG,BC,EF,GH,CD,DE,D
widely, the technology must meet several challenges. We
H} and the number aside from any vu E denotes the delay
describe the general functioning of unstructured P2P
networks, before discussing their drawbacks which form the of forwarding a message from v to u.
source of these challenges.

All Rights Reserved © 2012 IJARCET
159

ISSN: 2278 – 1323

Minimizing |p|, in turn, maximizes the broadcast scope of conclude that there exist some instances of small-world
disseminating a message, where the broadcast scope is networks that can match their physical network topologies.
defined as the number of distinct peers receiving the message Merugu et al. [5] do not detail how to create such a small-
with a specified TTL value. Notably, enlarging the broadcast world P2P network, however. In contrast, in our study, we
scope is particularly important for P2P search applications discuss how an unstructured P2P network that well matches
since this increases the probability of discovering a requested the physical network topology can be constructed. Our
object [14], [3], [15]. proposed algorithm is motivated by a rigorously analytical
In the following discussion, the message broad casting model.
protocol as mentioned above is called the scoped broadcast
(or broadcast for short) in this paper. Fig. 1 depicts an Liu et al. [14], [3], [15] present deterministic algorithms for
example of an unstructured P2P network that implements the the topology mismatch problem between unstructured P2P
scoped broadcast. networks and underlying networks. To optimize the P2P
Consider node A in the network. If TTL = 3, then the network topology (denoted by G), the studies [9], [10], [11]
messages originated by A will follow three communication suggest adding a new overlay link to G and removing an
paths, namely, p (A~D) = {ABCD; AGHD; AFED}, toward D. existing one from G iteratively such that the net operations
Among ABCD, AGHD, and AFED, the path ABCD has the can reduce the total delay cost of overlay links. While Liu et
minimal communication latency. That is, D accepts the al.’s solutions are elegant and require only local knowledge
message from the shortest path ABCD discards those from for each peer, their proposals provide no performance
paths AGHD and AFED. Here is 3, guarantee. In contrast, we present a novel proposal that is
driven by rigorous performance analysis. Unlike Liu et al.’s
and the scope of the broadcast message due to A is 7. algorithms, the latency of routing a scoped broadcast message
from any node v to another other node u in our presented
In this paper, we present an analytical model to show that
overlay network approximates the minimum (i.e., the end-to-
there exists unstructured P2P networks minimizing (1) for any
end delay from v to u), which is independent of the number of
pair of participating peers v and u. we propose a fully
nodes in the system. We compare our proposal to Liu et al.’s
decentralized algorithm based on the Self-Adaptive Topology
solution in extensive simulations, and the simulation results
Matching Method to construct the intended overlay networks.
validate the effectiveness of our proposal, showing that our
In our proposal, any peer in the system connects to as many as
design significantly outperforms Liu et al.’s.
possible of its geographically closest peers, subject to its
In order to address the limitations of the above cited work,
maximum number of connections. In addition, compared with
we propose a method, called SAT-Match that adaptively
distant peers, the peers in the proximity of the physical
changes the overlay structure of the P2P system to match the
network connect to one another in higher probability. We
underlying physical topology. By iteratively reducing the
show that our algorithm constructs the intended overlay
average logical link latency, the average response time of
networks in a lightweight fashion, and each participating peer
lookup routing is reduced. Moreover, it can be easily
implementing our algorithm requires only local knowledge.
implemented on existing infrastructures like CAN without
Reducing the average logical link latency can significantly introducing new operations. Each node maintains only a small
reduce the lookup response time, which requires an overlay number of neighbor states. Through intensive simulation
link to closely match its shortest IP route. Desired overlay experiments on large scale CAN overlays, we have shown the
construction algorithms correct the mismatch by selecting effectiveness of SAT-Match.
topologically close peers as neighbors. These algorithms need
to address several fundamental issues: (1) it should be 3. THE MISMATCH PROBLEM AND OUR
decentralized and scalable. (2) It should be accurate and APPROACH
adaptive to dynamically changing environments. (3) It should
be of low cost. We take a torus overlay structure in CAN system as an
example to explain our method, whose physical topology and
2. RELATEDWORK logical overlay are shown in Figures 1 and 2 respectively.
Starting from the overlay network in Figure 2, we show
Small-world networks exhibit low diameter [5], [8], [9]. By how the mismatch can be improved through self-adaptation. If
diameter, we mean the maximum hop count of routing a node A wants to route a message to node B, it needs to
message on the shortest path2 between any two nodes in a traverse the path A-C-B with the latency of 2+9+9 =20, or to
given graph network G=(V,E). Although a small-world traverse the path A-D-B with the latency of 12+3+9 =24, both
network G has a low diameter, the “delay” of routing a of which are measured by latencies of physical links in Figure
message between any two nodes in G may not be necessarily 1. However, node A could have routed the message to node B
small, considering that G is layered on top of a physical at the cost of 2 because there is a direct physical link with
network (e.g., the Internet). Without relying on rigorous latency of 2 between the two nodes. Because of the topology
performance analysis, Merugu et al. in their seminal study [5] mismatch, a query usually traverses some unnecessary links

160

ISSN: 2278 – 1323

before reaching its final destination. In Figure 2, after After probing in a TTL-k neighborhood and collecting a
changing the overlay structure on the left to the right one, the list of IP addresses, the source node uses a ping facility to
logical overlay can perfectly match the underlying physical measure the Round-Trip-Times (RTTs) to each of the nodes
topology. Compared to the two alternative routing latencies of that have responded. Then it sorts these RTTs and selects two
20 and 24 for node A to node B, the new routing latency is nodes with the smallest RTTs, based on which, the source
minimized to 2 node will select one zone associated with one of the nodes to
node in.
12 4 PERFORMANCE EVALUATIONS
D
A 4.1 Simulation Methodology and Performance Metrics

9 We have evaluated SAT-Match comprehensively through
2 simulation. The computing facilities are machines with Intel
core 3.06GHz processors and 2GB memory running windows
xp. As far as the logical overlay is concerned, we build our
B C own Java based CAN simulator by following the protocol
described in [1]. Each peer on the overlay is uniquely mapped
9 to one node in the IP layer. We choose CAN as the platform
because it adapts well to dynamic environments. However,
Fig 2. An example of a physical topology with 4 nodes. SAT-Match can also be easily deployed in other structured
P2P systems such as Chord and Pastry.
We have designed a method called SAT-Match to There are several parameters that can be tuned in the
adaptively change the overlay network connections following simulation.
the ideal topology matching scenario in Figures 1 and 2 for 1. D: the dimensionality of the Cartesian space
the purpose of reducing the average latency of logical hops. 2. K : the scale of the TTL-k neighborhood of one node
This iterative process completes until it is close enough to the 3. N: the size of the CAN system
zone where all its physically close neighbors are located and 4. topology type, which characterizes the node
no additional optimization is necessary. This continuously distribution on the transit domain and on leaf stub
adaptive topology matching process will achieve a global domains
topology matching optimization in a sufficiently large scope.
We define stretch as the ratio of the average logical link In our simulations, the number of nodes participating in the
latency over the average physical link latency to quantify the system is up to N =100000, and the default is 30,000. We
topology match degree, where we refer the logical link to the simulate a static environment in which the nodes participating
virtual link between a node and one of its direct neighbors in in the network do not join and leave. We also assess our
the overlay. proposal in a dynamic environment, and the peers in such an
environment have a lifetime with the exponential distribution
3.1 TTL-K Flooding and RTT Measurements of the mean equal to t minutes.
Since previous studies (e.g., [7], [12], [8]) concluded that
A recent study shows that flooding with a low number of the peers in an unstructured peer-to-peer network, represented
TTL hops is highly effective, which produces few redundant by G = (V, E), are randomly interconnected, we generate in
messages. The probing process of SAT-Match utilizes this the simulations Gs as random graphs using the algorithm
effective flooding. Having joined the system based on a DHT suggested in [8]. In G, each node v connects to at most M=6
assignment, a new node begins probing its neighborhood neighbors (i.e., the maximum number of connections v can
periodically. The source node floods out a message containing maintain). our proposal, each simulated peer in our design can
the source IP address the source timestamp and a small TTL also link to six peers at most. In particular, in our proposed
value k to all its neighbors on the overlay. Any node that overlay, each peer v can only connect up to three peers for it’s
receives this message responds to the source with its IP , while the size of vs is no more than 3. However,
address and decrements the TTL field in the messages by 1. If
unlike , v exploits its by rewiring the peers in an
the updated TTL value does not reach 0, this responding node
forwards the message to its neighbors. Since K is small, in the overlay network provided to our proposal. In the simulations,
end of this flooding, the message covers only a limited such a provided overlay network is a random graph in which
number of nodes in a small region. We define these nodes each node only connects to three neighbors at most.
being covered as the TTL-k neighborhood of the source node.
Specifically, the TTL-k neighborhood of a node refers to this 4.2 Effects of Varying the Number of Participating
node and all its direct neighbor nodes in the overlay. Peers

161

ISSN: 2278 – 1323

Corollary 2 states that our proposal minimizes the delay of knowledge, our proposal is the first design driven by the
sending a broadcast message between any two nodes v and u, rigorous analytical model, which approximates the optimum.
and approximates the minimum (i.e., the lower bound of In addition, our design guarantees the exponential
sending a message from v to u, ) independent of the broadcast scope. By exploiting the surrounding neighbors of
number of nodes participating in the system. We are thus peers with low communication delays, our overlay is
constructed to match the underlying network topology. Based
interested in the effects of varying the number of participating
peers in terms of broadcast delay in this section. on the concept, the resources of peers in our overlay can be
fully utilized.
Fig. 3 presents the simulation result for Random,
Random+THANCS, Hsiao, and Ours, given the end-to-end Furthermore, we will investigate more efficient and reliable
delays between peers due to BRIT E, by varying the number transmission schemes such as network coding and layered
coding to achieve high throughput, scalable and robust peer-
of nodes from 10,000 to 100,000. In Fig. 6, we measure which
is the averaged ratio of the delay of routing a broadcast to-peer streaming environments.
message from any node v to any other node u on the shortest
path in the overlay to that of sending the message from v to u ACKNOWLEDGEMENTS
in the physical network.
The authors would like to thank the Vice Chancellor, JNT
University, and Anantapur. Thanks are also due to Principal,
14 HOD and staff of Department of Computer Science and
Random+Thancs Engineering at JNTU College of Engineering, Anantapur for
12 Hsiao their support during the course of this work.
Noramalized delay

10 SAT

8 REFERENCES

6 [1] S. Ratnasamy, P. Francis, M. Handley, R. Karp, and
S.Shenker, “A Scalable Content-Addressable
4
Network,” Proc.ACM SIGCOMM ’01, pp. 161-172,
2 Aug. 2001
[2] X. Liao, H. Jin, Y. Liu, L.M. Ni, and D. Deng,
0 “Scalable Live Streaming Service Based on
1 2 3Number of 5
4 6 7 Interoverlay Optimization,” IEEE Trans. Parallel and
Nodes Distributed Systems, vol. 18, no. 12, pp. 1663-1674,
Fig. 3. The effect of varying the number of peers Dec. 2007.
[3] Y. Liu, “A Two-Hop Solution to Solving Topology
As Fig. 3 depicts, our proposal performs very well and Mismatch,” IEEE Trans. Parallel and Distributed
approximates the minimum delay of sending a message Systems, vol. 19, no. 11, pp. 1591-1600, Nov. 2008
between any two nodes. This validates the effectiveness of [4] H.-C. Hsiao, H. Liao, and C.-C. Huang, “Resolving
our design as motivated by our analytical result in Section 3 the Topology Mismatch Problem in Unstructured
and confirms Corollary 2. Second, while in [14], Hsiao et al. Peer-to-Peer Networks,” IEEE Trans. Parallel and
propose an overlay network with rigorous performance Distributed Systems, vol. 20, no. 11, pp. 1668- 1681,
guarantee in which the latency of routing a broadcast message http://doi.ieeecomputersociety.org/10.1109/TPDS.20
09.24
between any two nodes v and u is no more than in
[5] S. Merugu, S. Srinivasan, and E. Zegura, “Adding
expectation, what we present in this paper outperforms Hsiao, Structure to Unstructured Peer-to-Peer Networks:
and the delay of sending a message between v and u in our The Use of Small-World Graphs,” J. Parallel and
design is close to the minimum . Distributed Computing, vol. 65, no. 2, pp. 142-153,
Feb. 2005.
5. CONCLUSION [6] T. Qiu, G. Chen, M. Ye, E. Chan, and B.Y. Zhao,
“Towards Location-Aware Topology in
The major achievements of this work are related to the Both Unstructured and Structured P2P Systems,”
characterization of unstructured peer-to-peer architectures. Proc. 36th Int’l Conf. Parallel Processing (ICPP ’07),
Starting from classical analytical results, we introduced Sept. 2007
several novel topological models which put the emphasis on [7] S. Saroiu, P.K. Gummadi, and S.D. Gribble,
capturing the network dynamics, and in our view are very “Measurement Studyof Peer-to-Peer File Sharing
significant for peer-to-peer systems. To the best of our Systems,” Proc. Ninth SPIE/ACM Multimedia
Computing Networking (MMCN ’02), Jan. 2002

162

ISSN: 2278 – 1323

[8] S. Jin and H. Jiang, “Novel Approaches to Efficient
Flooding Search in Peer-to-Peer Networks,”
Computer Networks, vol. 51,no. 10, pp. 2818-2832,
July 2007.
[9] A. Medina, A. Lakhina, I. Matta, and J. Byers,
“BRITE: An Approach to Universal Topology
Generation,” Proc. Ninth Int’l Workshop Modeling,
Analysis, and Simulation of Computer and
Telecomm. Systems (MASCOTS ’01), pp. 346-353, [14] Y. Liu, L. Xiao, and L.M. Ni, “Building a Scalable
Aug. 2001 Bipartite P2P Overlay Network,” IEEE Trans.
[10] Hung-Chang Hsiao “A Near-Optimal Algorithm Parallel and Distributed Systems,vol. 18, no. 9, pp.
Attacking the Topology Mismatch Problem in 1296-1306, Sept. 2007.
Unstructured Peer-to-Peer Networks “IEEE
Transactions ON Parallel and Distributed
System”,Vol .21 No.7, July 2010 K L Haritha received the B. Tech Degree in computer
[11] K.P. Gummadi, R.J. Dunn, S. Saroiu, S.D. Gribble, science and Engineering from JNT University, Anantapur,
H.M. Levy, and J. Zahorjan, “Measurement, India, in 2009. She is currently pursuing the M.Tech degree in
Modeling, and Analysis of a Peer-to-Peer File- computer science with JNT University, Anantapur, India.
Sharing Workload,” Proc. 19th ACM Symp. She research interests include the areas of computer
Operating Systems Principles (SOSP ’03), pp. 314- networking, including mobile ad hoc networks.
329, Oct. 2003.
B Lalitha received the B.Tech degree in computer
[12] G.Pandurangan, P. Raghavan, and E. Upfal, science from SITAMS Chittoor,JNTUH, India, in 2003. And
“Building Low-Diameter Peer-to-Peer Networks,” M.Tech degree in computer science from JNT University,
IEEE J. Selected Areas in Comm., vol. 21, no. 6, pp. Anantapur, India, in 2005. She is currently working toward
995-1002, Aug. 2003. the Ph.D degree in computer science with JNT University,
[13] Y. Liu, L. Xiao, X. Liu, L.M. Ni, and X. Zhang, Anantapur, India.
“LocationAwareness in Unstructured Peer-to-Peer She is currently an Assistant Professor of computer
Systems,” IEEE Trans.Parallel and Distributed science and engineering with JNTUA College of Engineering,
Systems, vol. 12, no. 2, pp. 163-174, Feb. 2005. Anantapur, India. Her research interests include the areas of
Distributed Computing, Peer-to-Peer Networks.

163

159 163

More Related Content

What's hot

Viewers also liked

Similar to 159 163

More from Editor IJARCET

Recently uploaded

159 163