Routing performance of structured overlay in Distributed Hash Tables (DHT) for P2P

Bulletin of Electrical Engineering and Informatics
Vol. 8, No. 2, June 2019, pp. 389~395
ISSN: 2302-9285, DOI: 10.11591/eei.v8i2.1449  389
Journal homepage: http://beei.org/index.php/EEI
Routing performance of structured overlay in Distributed Hash
Tables (DHT) for P2P
Rafiza Ruslan1
, Ayu Shaqirra Mohd Zailani2
, Nurul Hidayah Mohd Zukri3
, Nur Khairani
Kamarudin4
, Shamsul Jamel Elias5
, R. Badlishah Ahmad6
1,2,3,4
Department of Computer Science, Universiti Teknologi MARA Perlis, Perlis, Malaysia
5
Universiti Teknologi MARA (UiTM) Kedah, Malaysia
6
Faculty of Informatics and Computing, Universiti Sultan Zainal Abidin (UniSZA), Malaysia
Article Info ABSTRACT
Article history:
Received Oct 16, 2018
Revised Dec 15, 2018
Accepted Mar 12, 2019
This paper presents a routing performance analysis of structured P2P overlay
network. Due to the rapid development and hectic life, sharing data
wirelessly is essential. P2P allows participating peers move freely by joining
and leaving the network at any convenience time. Therefore, it exists
constraint when one measuring the network performance. Moreover, the
design of structured overlay networks is fragmented and with various design.
P2P networks need to have a reliable routing protocol. In order to analyse the
routing performance, this work simulates three structured overlay
protocols-Chord, Pastry and Kademlia using OMNeT++ with INET and
OverSim module. The result shows that Pastry is the best among others with
100% routing efficiency. However, Kademlia leads with 12.76% and 18.78%
better than Chord and Pastry in lookup hop count and lookup success latency
respectively. Hence, Pastry and Kamelia architectures will have a better
choice for implementing structured overlay P2P network.
Keywords:
Chord
DHT
Kademlia
P2P
Pastry
Copyright © 2019 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Rafiza Ruslan,
Department of Computer Science,
Universiti Teknologi MARA Perlis, Perlis, Malaysia.
Email: rafiza.ruslan@perlis.uitm.edu.my
1. INTRODUCTION
Peer-to-peer (P2P) systems are widely used for data sharing that exploits a self-organization for
optimizing resource allocation. P2P is an example of overlay networks that implements a network service
that is not available in the physical network. In a simplest form, P2P network is connected between two or
more PCs in a decentralized environment similar to Mobile Ad-Hoc Network (MANET) that consist of nodes
communicate each other without any fixed infrastructure through the wireless devices [1]. P2P overlay
networks are divided into two classes which is unstructured and structured [2]. Unstructured P2P networks
use flooding search protocols where the demand is communicated to all companions inside a specific range
with Time to Live (TTL) instrument and also have no precise control over resource emplacements such like
Gnutella. Other examples of application file sharing that apply P2P system are Napster, Morpheus, Freenet,
BitTorrent and Skype. On the other hand, structured P2P network refers to the tightly controlled topology
that is maintain over a network graph where resources are not placed randomly on nodes but in a
deterministic manner, i.e. using Distributed Hash Table (DHT). P2P need to have better structured overlay
algorithm for P2P due to the increasing number of users in P2P network [3]. DHT is one of the distributed
memory method where it distributes the contructions of hash table [4]. In DHT based system, identifier (Id)
for each peer and item is generated using hash function. The mapping of Ids into set of peers of structure
overlay is done by DHT in a distributed fashion; called routing or lookup [5]. Structured overlay network
protocol that use DHT are Chord, Pastry, Kademlia etc [6].

 ISSN: 2302-9285
Bulletin of Electr Eng and Inf, Vol. 8, No. 2, June 2019 : 389 – 395
390
Chord is the first DHT-based structured P2P overlay protocol [7]. It uses consistent hashing and
hashes each peer and key into a same m-bit circular identifier (Id) space, where m is a system parameter [7].
Chord peer uses routing tables to route the lookup message to the destination peer in O(log N) complexity,
where N is total number of peers. Chord uses periodic stabilization mechanism to correct the routing table
periodically. Pastry organizes each peers in 128-bit circular Id space that uses prefix matching to route the
messages in O(logb N) complexity, where N is total number peers and b is system parameter [8].
Neighbouring peers periodically exchange keep-alive messages in order to maintain the routing table.
Finally, Kademlia organizes each peer by assigning 160-bit Id and the each peer contains list of entries
known as routing buckets [6]. Kademlia uses XOR based closeness algorithm to calculate the closest peers,
then sends parallel lookup messages to multiple buckets at complexity of O(logb N), where N is total number
of peers and b is system parameter. Kademlia peer employs maintenance algorithm which periodically
checks the routing buckets through which there has not been a lookup since last stabilization. Kademlia only
ensures that at least one entry in each bucket is alive since last stabilization, where as Chord and Pastry
ensures all routing entries are alive and up to date.
Structured overlays use its corresponding routing algorithm and the routing table to route the
messages to the destination peer. The random join and leave of peers in P2P system causes the routing table
to be incorrect and out of date. This routing timeout may results in lookup failure. The rate of failure is
dependent on the rate of peer join and leave or known as mean lifetime of the peers. Each structured overlay
uses the maintenance algorithm to keep the routing table up_to_date. The correctness of the routing table
depends on how frequently the routing table is maintained. Hence, the routing performance increases with the
increase in frequency of maintenance. The asymptotic analysis of routing performance of the structured
overlay protocols - Chord, Pastry and Kademlia are logarithmic. This paper present the analysis of routing
performance on structured overlay in DHT which are Chord, Pastry and Kademlia to find better structured
overlay for P2P network. In this study, the analysis is according to the performance metric in terms of routing
efficiency, lookup hop count and lookup success latency. However, due to the simulation resource limitation,
we only simulate 100-500 peers of P2P networks and range of 100-1000 sec of mean lifetime.
There are also previous related works that compare the structured DHT overlay network. In [5], the
routing performance comparison are also among Chord, Pastry and Kademlia, where only routing efficiency
is being considered. Routing efficiency is calculated as the ratio of successfully routed messages to total
number of messages sent. The simulation increases with mean lifetime of peers. The result shows that
Kademlia has the best routing efficiency followed by Pastry and Chord. A simulation carried out by [9] on
Chord, Pastry, Kademlia, Broose and EpiChord in mobile networks having high churn rate. From the result,
Kademlia are also chosen to be the best P2P overlay to implement on the mobile nodes based on routing
efficiency. Due to the importance of lookup latency and routing efficiency, [10] developed a mathematical
model that combine these two metrics. They found that value of effective latency using the mathematical
model matches their simulations. This reason motivates to simulate the other performance metric which is the
workload per node or the lookup hop count. Thus, this work analyzed three performance metrics which are
routing efficiency, lookup success latency and lookup hop count.
The remainder of the paper is organized as follows. Section 2 illustrate the research method on
simulating Chord, Pastry and Kademlia. Section 3 demonstrates the results and analysis on the routing
performance according to the experiments. Finally, the whole paper is concluded in Section 4.
2. RESEARCH METHOD
This work simulates structured overlay protocols using OMNeT++ with INET [11] and OverSim
module. OverSim is an open-source P2P simulation framework that has 3-tier architecture designed using the
NED language [3]. OverSim has been described as a powerful tool for evaluating and developing P2P
applications [12, 13]. Figure 1 shows the 3-tier architecture consists of three layers such as application layer,
overlay layer and underlay layer.
In application layer, key-based routing (KBR) is implemented where it periodically sends test
messages to random nodes and records different parameters such as number of lookup, message sent, the
number of lookup failed, messages delay and hop-counts [8]. In the same layer, P2P Apps for example DHT
TestApp is used to test the DHT application, by performing DHT PUT and GET requests with randomly
generated data items [14]. Important to realize, these applications are implemented as modules and interface
with overlays through KBR API, which represent basic capabilities common to all structure overlays. In the
overlay layer, OverSim provides a number of different network models, for both structured (Chord,
Kademlia, Pastry, Broose and Koorde etc.) and unstructured overlays (GIA etc) [14]. In the underlay
network, the INET framework can be used for routing protocol implementation and also support wireless and

Bulletin of Electr Eng and Inf ISSN: 2302-9285 
Routing performance of structured overlay in Distributed Hash Tables (DHT) for P2P (Rafiza Ruslan)
391
mobile simulation as well [5]. In addition, OverSim provides multiple underlay models to allow detail
simulation; and provides single host underlay for simulation of a single node.
Figure 1. 3-tier simulation framework implemented in OverSim simulator [3]
In this simulation, the routing efficiency, lookup hop count and lookup success latency of the peers
are calculated and recorded. The data presented in section 3 is the mean of 3 independent experiments. The
following performance metrics have been analyzed in order to measure the performance [15].
Routing Efficiency: Routing efficiency also known as lookup success rate is calculated by taking
the ratio of total number of lookup messages managed to succees before timeout and the total number of
lookup messages broadcastes by the system. This metric is calculated using (1).
(1)
Lookup Hop count: Hop count or workload per peer is calculated by taking the ratio of total
broadcasted complex query messages and total number of nodes required to successfully forward the
messages to the destination node. This is defined in (2).
(2)
Lookup Success Latency: Lookup success latency is defined as the time taken for a packet of data
to get from one designated point to another and sometimes measured as the time required for a packet to be
returned to its sender.
This work simulates the routing performance for different network size and mean lifetime of the
peers as mentioned in Table 1. In this simulation, the number of peers were limit to 500. The mean lifetime
of the peers are varied between 100 sec to 1000 sec. This limitation is due to the limited resources for
executing the simulation.
Table 1. Parameters used for simulation
Name of the Parameters
Architecture
Chord Pastry Kademlia
Number of peers 100, 200, 300, 400, 500 100, 200, 300, 400, 500 100, 200, 300, 400, 500
Mean life-time of the peer (sec) 100, 300, 500, 700, 1000 100, 300, 500, 700, 1000 100, 300, 500, 700, 1000
3. RESULTS AND ANALYSIS
3.1. Routing efficiency
Two scenarios has been conducted to collect the routing efficiency results. First experiment which is
increasing network size from 100 to 500 peers with the fixed mean lifetime of 1000 sec. Figure 2 illustrates
the routing efficiency vs network size.

 ISSN: 2302-9285
392
Figure 2. The results on Chord, Pastry and Kademlia based on routing efficiency vs network size
Chord has the lowest routing efficiency over network size among Pastry and Kademlia. The lowest
routing efficiency ratio is Chord where the ratio is 0.8579 at 100 peers and gradually decrease when the
network size increase to 500 peers. Pastry maintained the routing efficiency to 1.0 until network size 500
peers then drop to 0.9998. Kademlia has a slightly lower ratio compared to Pastry. If the routing efficiency is
1.0, meaning that it is a good routing with high network stability. Chord organizes its routing table in
single-dimensional space, whereas Pastry and Kademlia have bigger routing table and organized its routing
table in two-dimentional space. Second scenarios for routing efficiency are increasing mean lifetime with the
fixed network size (100 peers). Figure 3 illustrates the routing efficiency vs mean lifetime.
Figure 3. The results on Chord, Pastry and Kademlia based on routing efficiency vs mean lifetime
Based on Figure 3, the results are similar where, Chord has the lowest compared to Pastry and
Kademlia algorithm. The ratio is 0.0959 (Chord algorithm) on 100 seconds means lifetime. The ratio in
Chord is linearly increased as the mean lifetime increases. Kademlia has a slightly drop compared to Pastry.
The highest routing efficiency is the ratio of 1.0 which Pastry on 100 seconds to 1000 seconds mean lifetime.
This proves that Pastry are the best algorithm for routing efficiency.
3.2. Lookup hop count
In lookup hop count, the similar scenarios were conducted. First, increasing network size from 100
to 500 peers with fixed mean lifetime 1000 sec. Figure 4 illustrates the lookup hop count vs network size.

393
Figure 4. The results on Chord, Pastry and Kademlia based on lookup hop count vs network size
In Figure 4, lookup hop count for Pastry shows a constant ratio of 0 where the values were too
small. This is due to the roundup value process. Between Kademlia and Chord, Kademlia has the better
lookup hop count at 100 peers, where lookup hop count is 1.2763. Chord has the ratio of 3.1234. When the
network increases to 500 peers, both mean lookup hop count for Kademlia and Chord increased respectively.
The result shows Pastry has the lowest lookup hop count compared to Chord and Pastry. The lower the
lookup hop count, the better the routing protocol. This result replicates the previous routing efficiency result,
where Pastry has ratio of 1.0 routing effiency. Second simulation is increasing mean lifetime from 100 to
1000 sec with fixed network size of 100 peers. Figure 5 illustrates the lookup hop count vs mean lifetime.
Figure 5. The results on Chord, pastry and kademlia based on lookup hop count vs mean lifetime
Based on Figure 5, Pastry has similar results with lookup hop count vs network size, where the
lookup hop count is near to zero. Chord has the highest lookup hop count. Therefore, Pastry shows better
routing in lookup hop count among other algorithms.
3.3. Lookup success latency
Similar lookup success latency, two scenarios have been conducted with increasing network size
from 100 nodes to 500 nodes with the fixed mean lifetime 1000s. Figure 6 shows the plotted graph for the
lookup success latency vs network size.

 ISSN: 2302-9285
394
Figure 6. The results on Chord, Pastry and Kademlia based on lookup success latency vs network size
Lookup success latency is where latency time takes for data packets to be stored or retrieved. The
lower lookup success latency are the better in the routing performance. As illustrated in Figure 6, Chord has
the highest lookup success latency and Kademlia has the lowest lookup success latency. Second simulation
were the increasing mean lifetime with the fixed network size of 100 peers. Figure 7 illustrates the lookup
success latency vs mean lifetime. Chord has the highest lookup success latency compared to Pastry and
Kademlia. Kademlia has the lowest lookup success latency.
Figure 7. The results on Chord, Pastry and Kademlia based on lookup success latency vs mean lifetime
This work simulates three performance metrics which are routing efficiency, lookup hop count and
lookup success latency. The parameter setting for network size is from 100 to 500 peers and the mean
lifetime is from 100-1000 sec. This set of simulation is considered a small-scale of network environment due
to the simulation resource limitation. The result shows that Pastry has the highest ratio in routing efficiency.
The results are not similar to the previous research by [5]. However, the difference of routing efficiency
between Pastry and Kademlia is not significant. It is assume that if the simulation executes in a large scale of
network and in a higher mean lifetime, Kademlia has the possibility to obtain highest ratio in routing
efficiency similar to the result obtained by [5]. Chord has the lowest routing efficiency due to Chord
organizes its routing table in single-dimensional space, whereas Pastry and Kademlia have bigger routing
table and organized its routing table in two-dimentional space. Lookup hop count is defined as the workload

395
per peer in the network [15]. A low ratio of lookup hop count is good for routing where the workload per
peer is low as well. The results for lookup hop count are that Pastry has the lowest ratio, followed by
Kademlia and Chord. The value obtained for Pastry from the simulation is rounded to 0. Here, the results for
lookup hop count are parallel to the results in routing efficiency for Pastry in this work. Finally, the last
simulation shows that Kademlia obtained the lowest lookup success latency ratio, followed by Pastry and
Chord. The lookup success latency between Kademlia and Pastry also has not much difference.
4. CONCLUSION
In conclusion, this project was completed and achieved its objective. The main objective of this
research was to simulate the structured overlay protocol Chord, Pastry, and Kademlia that use DHT using
OverSim Simulator using NED language. The second objective is to evaluate three DHT-based structure
overlays which are Chord, Pastry and Kademlia on the network performance in terms of routing efficiency,
lookup hop count and lookup success latency for P2P. The simulations have been divided into two scenarios,
the first scenario was created according to the increasing number of nodes and the second scenario were
created according increasing mean lifetime. For the first scenario in term of routing efficiency Pastry has the
best routing efficiency is 1.0. Next, in term of lookup hop count and lookup success latency Kademlia is the
best algorithm when has the lowest lookup hop count and lookup success latency. For the second scenario,
performance measure in term of routing efficiency the best algorithm is Pastry but not have much different
with Kademlia which slightly drop from Pastry. For lookup hop count and success latency Kademlia has the
lowest for these two performance measurements. This means Kademlia is the best algorithm among Pastry
and Chord.
ACKNOWLEDGEMENTS
We thank Dr. Saiful Khan from University of Oxford’s e-Research Centre and Dr. Mr. Mohammad
Elsheikh from Universiti Malaysia Perlis for the useful discussions, guidance, and assistance in this research.
REFERENCES
[1] M. Caleffi and L. Paura, “P2P over MANET: Indirect tree-based routing,” in 7th Annual IEEE International
Conference on Pervasive Computing and Communications, PerCom 2009, 2009, pp. 18–22.
[2] S. A. Abid, “An Application Of 3D Logical Structure In DHT Paradigm For Efficient Communication In
MANET-PhD Thesis,” Unversity of Malaya, Kuala Lumpur, 2014.
[3] J. Leng and T. Li, “Research on P2P Network Resource Search Model,” in International Conference Advanced
Engineering and Technology Research (AETR 2017) Research, 2018, vol. 153, no. Aetr 2017, pp. 55–60.
[4] C. Yan, “Distributed hash table based routing algorithm for wireless sensor networks,” in Proceedings-2014 5th
International Conference on Intelligent Systems Design and Engineering Applications, ISDEA 2014, 2014,
pp. 430–433.
[5] S. Khan, A. Gani, and M. Sreekandath, “The routing performance of logarithmic-hop structured P2P overlay,” 2011
IEEE Conf. Open Syst. ICOS 2011, pp. 208–213, 2011.
[6] P. Maymounkov and D. Mazi, “Kademlia : A Peer-to-peer Information System Based on the XOR Metric,”
PeertoPeer Syst., pp. 53–65, 2002.
[7] I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan, “Chord: A Scalable P2P Lookup Service for
Internet Applications,” Proc. of SIGCOMM, pp. 1–23, 2001.
[8] S. Khan, A. Gani, S. Raviraja, and Shashikantha, “A study of broadcasting over structured overlay,” 2015 7th Int.
Conf. Commun. Syst. Networks, COMSNETS 2015-Proc., 2015.
[9] F. Chowdhury, “Performance Evaluation of Structured Peer-to-Peer Overlays for Use on Mobile Networks,” 2014.
[10] N. Varyani, S. Nikhil, and V. S. Shekhawat, “Latency and routing efficiency based metric for performance
comparison of DHT overlay networks,” in Proceedings-IEEE 30th International Conference on Advanced
Information Networking and Applications Workshops, WAINA 2016, 2016, pp. 337–342.
[11] OMNeT++, INET Framework for OMNeT++. 2012.
[12] J. Moorhouse, L. Liu, Z. Li, and Z. Ding, “Modelling and Simulation of Peer-to-Peer Overlay Network Protocols
using OverSim,” 2013 UKSim 15th Int. Conf. Comput. Model. Simul., pp. 144–149, 2013.
[13] T. Kunz, B. Esfandiari, S. Ngozi, and F. Ockenfeld, “P2P Overlay Performance in Large-Scale MANETs,” Int. J.
Commun. Netw. Syst. Sci., pp. 147–173, 2018.
[14] J. Furness, M. Kolberg, and M. Fayed, “An evaluation of chord and pastry models in OverSim,” Proc.-UKSim-
AMSS 7th Eur. Model. Symp. Comput. Model. Simulation, EMS 2013, pp. 509–513, 2013.
[15] S. Khan, “Complex Searching Over Structured Overlay Under Dynamic Environment-Master Thesis,” University
Of Malaya, Kuala Lumpur, 2011.

Routing performance of structured overlay in Distributed Hash Tables (DHT) for P2P

More Related Content

What's hot (18)

Similar to Routing performance of structured overlay in Distributed Hash Tables (DHT) for P2P (20)

More from journalBEEI (20)

Recently uploaded (20)

Routing performance of structured overlay in Distributed Hash Tables (DHT) for P2P