Peer-to-Peer Networking and Applications (2021) 14:888–897

https://doi.org/10.1007/s12083-020-01049-3

Trust estimation in peer-to-peer network using BLUE

Ruchir Gupta1 · Yatindra Nath Singh2 · Antriksh Goswami3

Received: 2 March 2020 / Accepted: 11 December 2020 / Published online: 11 January 2021
© The Author(s), under exclusive licence to Springer Science+Business Media, LLC part of Springer Nature 2021

Abstract
Free-riding is a major problem in peer to peer networks. One of the ways to resolve this problem is by using a reputation
management system. Each node provides resources to nodes with higher reputation, as he himself gains in reputation in
proportion to the receiver’s reputation. So, free-riding becomes a non-optimal strategy as there is no resulting reputation gain,
and hence no available resources over time. An efficient and accurate trust estimation method in the reputation management
system is needed for effectiveness of this strategy. The existing trust estimation methods generally do not consider the
uncertainties in the inputs. In this paper, we propose a new trust estimation method using BLUE (Best Linear Unbiased
Estimator) that also takes into account the uncertainties in the input variables. The results show that using the proposed
estimation method, effective trust estimation is achieved in the peer-to-peer resource sharing systems. Thus the free-riding
can be countered more effectively.

Keywords Reputation · BLUE · Free-riding

1 Introduction to the community by contributing some resources. These

Peer-to-peer systems have attracted considerable attention
recently as these systems are more scalable than the
client-server systems. However, the distributed nature of
peer-to-peer networks brings in many challenges for the
system designers. These networks are generally designed
considering the fact that every node in the system is
honest and cooperative. It means, that if a node takes some
resource from the community, it will also reciprocate back
 Ruchir Gupta
rgupta.cse@itbhu.ac.in
Yatindra Nath Singh
ynsingh@iitk.ac.in
Antriksh Goswami
antriksh goswami@iiitvadodara.ac.in
1
Department of Computer Science and Engineering, Indian
Institute of Technology (BHU), Varanasi, 221005, India
2
Electrical Engineering Department, Dean of Infrastructure
and Plannning, Chairman CEMMC, Indian Institute
of Technology, Kanpur, U.P.-208 016, India
3 ratings, reputation is decided. This reputation score helps
Computer Science & Engineering Department, Indian
Institute of Information Technology Vadodara, Block-9,
GEC, Sector - 28, Gandhinagar, 382028, India
transactions, happening independently, tend to balance
each other over a long duration in a stable system. In
real systems, nodes behave in a selfish manner and try
to maximise their utility. This selfishness may result in
their non-cooperative behaviour unless appropriate control
mechanisms are implemented. This phenomenon is best
explained by famous Prisoners’ Dilemma, in which Nash
Equilibrium (NE) [1] is achieved when both prisoners
deceive each other. Similarly, considering nodes in a file-
sharing network as players, it turns out that none of the
players shares the resources under Nash equilibrium [2].
This tendency of nodes to draw resources from the network
and not giving anything in return is termed as ‘Free-Riding’.
An Experimental study [3] on the Gnutella network in 2005
also confirmed free-riding by showing that the number of
free-rider nodes in the network is as high as 85%. Such
type of problems also exists in e-commerce systems like e-
bay. In e-commerce portals like e-bay, buyers and sellers
generally do not know each other and, hence the possibility
of cheating or providing a product or service of inferior
quality always exists. To avoid this, e-bay uses a rating
based reputation system. After every transaction, the user
gives a feedback rating to his counterpart and based on these
the buyer to decide about carrying out the transaction (http://
www.ebay.com).
Peer-to-Peer Netw. Appl. (2021) 14:888–897
Reputation management systems are also employed in
peer-to-peer networks to overcome the problem of free-
riding as well as to ward off different types of attacks.
Trust or reputation is a central quantity in a reputation
management system. It helps a node in predicting the
behaviour of any other node whom it is going to interact
with and hence to minimise the possibility of misbehaviour.
In a peer to peer file sharing network, trust or reputation of
a node measures its co-operative behaviour towards other
nodes. A node seeking some resource from another node
measures the ratio of received resource to the requested
resource after every transaction and updates the trust
value accordingly. Hence, the performance of a reputation
management system depends upon the methods for trust
estimation.
A simple way to estimate the trust is the ratio of
received resources to requested ones. This simple method
has following issues:
1. In peer-to-peer networks, peers usually ask for an
excessive amount of resources so that their actual
demand, which is lesser in amount, can be fulfilled.
The requester refuses the extra resource offered and this
extra resource does not contribute to the reputation of
serving node.
2. Once the requesting node decides about the node from
which it is going to take data, both of the nodes
decide about the data rate according to their upload and
download capacities to the nearest routers. But at the
same time, the underlying network may not be able to
provide the agreed data rate because of congestion in
the network at different routers. So, even if the service
provider node is willing to give data, it may not be
able to send it at the committed rate due to limitations
of the underlying transport network. This affects the
reputation assignment as reputation is assigned based
on the actual data rate.
3. If some node has already got too many requests, it will
not be able to provide the quality of service that it could
have provided with the lesser load. Due to this fact,
the requesting node should estimate the reputation of
the service provider node considering the load and the
previous transactions with that node to avoid too many
requests being made to it.
In this paper, we are proposing a trust estimation method
using BLUE (Best Linear Unbiased Estimator) [4] which
resolves the aforementioned issues and requires almost the
same amount of memory and computation. The contribution
of the paper is as follows.
1. We propose a generalized formula for calculating the
willing service rate by any service provider.
889
2. We calculated the uncertainty in the offered service rate
due to load variation at the service provider.
3. An effective trust estimation formula is proposed
considering the actual service rate, willing service rate,
feasible service rate based on the network congestion,
and requested service rate.
The remainder of this paper is organised as follows.
Section 2 discusses the related work in reputation man-
agement, and Section 3 describes the system model. In
Section 4, issues in trust estimation have been discussed.
Proposed method explaining the estimator for trust using
BLUE is derived in Section 5. Finally, Section 6 presents
the numerical results and Section 7 concludes the paper.
2 Related work
Trust is directly proportional to the extent of cooperative
behaviour of a peer. It can be measured as the contributions
made by a serving node [5–8] or as the quality of service
obtained from the serving node [6, 9–15]. It can also be
measured as the ratio of the resources received and the
resources provided by a node [16–18].
One more approach is to calculate the ratio of the sum
of resources received, to the resource requested by the
node for the last ten transactions [19]. Further, in [20],
provider’s inbound bandwidth and requester’s reputation
have been taken as input to a fuzzy logic method to calculate
the incentives for sharing. This method also provides an
incentive to avoid overloading problems for initial file
providers. Suretrust [21] proposed the trust model for two
level p2p networks. Each subnet has one superpeer which
is responsible for calculating trust of the peer in the subnet.
The trust of the superpeer is correlated with the trusts of the
peers in the subnet.
Eigen-Trust [12] uses the sum of positive and negative
ratings, Peer-Trust [13] normalises the rating on each
transaction, whereas Power-Trust [14] uses the Bayesian
approach to locally calculate the reputation.
Mengshu et al. [22] took the ratio of successful
transactions to total transactions. PET [23] categorises
services qualities of different transactions into four types
and then compute the total average of a node by giving
different weight to each category of transaction. In [24],
different reputation scores have been calculated for different
types of resources. In Fuzzy-Trust [25], nodes performed
fuzzy inference on the parameters to locally calculate the
trust score for another node ‘x’ and then aggregate it with
trust scores of the node ‘x’ as received from other nodes
using their weights. In [26], peers maintain a binary vector
of m bits. After a transaction, one or zero is added at
the most significant position of the vector after shifting
890
right all the previously placed bits once. This trust vector
is considered as an m bit binary number. To compute the
trust, m bit binary number is divided by 2m . This ensures
that the trust value lies between 0 and 1. But none of the
above works considers the uncertainties in the input while
estimating the trust value. Recently trust estimation is used
for providing the services by one application to another.
OPRE [27] and SCARE [28] are the methods proposed for
trust estimation in Service-oriented architecture.
Much recently, trust is also being measured and used in
the P2P based applications like blockchain, social networks
etc. An order-driven trading service between a manufacturer
and a customer in the blockchain-based Industrial IoT
architecture is proposed in [29]. Authors measure the
reputation of the manufacturer by taking into account the
credibility and capability of the manufacturer. The proposed
method is customized to increase reliability and accuracy. A
blockchain based publisher-subscriber system is proposed
in [30]. The publisher publishes the topic on the blockchain
and subscriber adds his interests. After topics of interest
match, the publisher transfers the encrypted content to the
blockchain and subscriber obtains the content by decrypting
the content. The subscriber marks the reputation of the
publisher based on the content. A new reputation-based
consensus protocol for blockchain is proposed in [31]. In
this work, the reputation of a node is calculated based on
its assets, transaction activity, and consensus participation.
In [32], more weightage is given to the reputation provided
by a more trustworthy peer, i.e. a peer having high
global reputation, in reputation aggregation process. It
helps in identifying peers as good peers or malicious
peers.
Fig. 1 System architecture
Peer-to-Peer Netw. Appl. (2021) 14:888–897
Further, Paper [33] showed that the reputation could be
distorted by the peers who mutually boost the reputation of
each other.
3 System model
In this paper, we are considering a peer-to-peer network in
which there are no dedicated servers. Peers in this network
are rational, i.e. they are only interested in their welfare.
An access link connects them to a backbone link forming
the network as shown in Fig. 1. We are assuming that the
network is heavily loaded, i.e. every peer has a sufficient
number of pending download requests. Hence these peers
are contending for the available transmission capacity. We
also assume that every peer is paying the cost of the access
link as per the use. So, every peer wants to maximise its
download capacity and minimise its upload capacity so that
it can get maximum utility for its spending. This leads to the
problem of free-riding.
This paper deals with the resource allocation part only
and resource discovery is out of the scope of this paper. Still
for sake of clarity, we consider that each peer has a table
which contains the IP addresses of the neighbouring nodes.
Whenever a peer is unable to fulfil its requirement from its
neighbouring nodes, it makes a query request. As a result of
query resolution, peer receives the IP address of the peer that
has the required content. Then, it directly sends request to
the owner, given that the resource transfer can happen from
any peer to any other peer in the network.
Further, every node maintains a reputation table as shown
in Fig. 1. In this table, a node maintains reputations of the
Peer-to-Peer Netw. Appl. (2021) 14:888–897
nodes with which it has interacted. Whenever it receives
a resource from any node, it adjusts the reputation of that
node accordingly. When a node asks for the resource from
this node, it checks the reputation table, and according to
the reputation value of the requesting node, it allocates
the resource to that node. This ensures that every node is
facilitated from the network as per its contribution to the
system and free-riding is discouraged.
We model the progress of our system in the periods of
equal and fixed time units. Peers decide the data service rate
that it is willing to contribute for this period. We consider
qw,j i [n] as the willing service rate from peer j to peer i for
the time period n.
4 Issues in trust estimation
In a stable peer to peer networks, the download should be
equal to upload for a node [34]. It looks like there is no
gain in this condition. But, even in this scenario, the node
gains due to interaction with others which leads to wider
availability of content along with a good quality of service
because of the increase in reputation of the node. Further,
in any society, even when resources are given and taken are
the same for each participating entity, they have a better
chance of survival compared to a community where there is
no interaction. Thus, the interaction itself is an incentive for
all the peers as it increases their utility.
A node will usually try to get the content and avoid
uploading it as this maximises gain for it. Therefore,
free-riding becomes an optimal strategy. So a reputation
management system needs to be enforced to safeguard the
interest of every node by controlling the free-riding.
In reputation management system, every node maintains
a reputation table. In this table, the node maintains the
reputations of the nodes with which it has interacted.
Whenever it receives a resource from any node, it adjusts
the reputation of that node accordingly. When a node asks
for the resource from this node, it checks the reputation
table, and according to the reputation value of the requesting
node, it allocates the resource to that node. This ensures
that every node is facilitated from the network as per its
contribution to the system and consequently, free riding is
discouraged.
For using such a reputation management system, a node
needs to estimate the trust value of the nodes interacting
with it. This estimation can be made based on requested
and actual transfer rates and other parameters after every
transaction. As shown in Section 2, estimation of trust has
been done by a number of ways in the literature. We have
891
modified one such method where a node estimates the trust
by
Zijk
tijk = . (1)
Rijk
Here Rijk represents the amount of resources requested
by node i from node j for k th transaction; Zijk represents the
amount of resource received by node i from node j in k th
transaction. In other methods also, the uncertainties have not
been generally considered. So they can also be modified for
estimating trust from the observables.
In the estimation of trust value by such a method, a few
important points are missed out. These points are as follows.
1. In peer-to-peer networks, when a peer asks for some
resource (i.e. data rate), it is not guaranteed that it will
get the asked resource. So, generally, the peer asks for
a larger amount of resource than needed, and when it
is offered more resource than its requirement, it refuses
the extra offered resource. Also, it does not give any
credit for this extra offer.
2. Once requesting node decides about the node from
which it is going to take data, both of the nodes
decide about the rate of data according to their upload
and download capacities. But at the same time, the
underlying network may not be able to provide the
agreed data rate because of congestion in the network
at different routers. So, even if the service provider
node is willing to give data, it may not be able
to send at the committed rate, due to limitations
of the underlying transport network. This affects the
reputation assignment as reputation is assigned based
on the actual data rate.
3. If some node has already got too many requests, it will
not be able to provide the quality of service that it could
have provided with the lesser load. Due to the above
fact, the requesting node should estimate the reputation
of the service provider node considering the load and
the previous transactions with that node to avoid the
distortion.
5 Proposed method
To resolve the above issues, node will use Eq. 8 for
calculation of trust value (tij ).
Let the data rate qr,ij is requested by the node i. At the
serving node j , qw,j i , the willing service rate from node
j to node i, is decided according to point 3 mentioned
892 Peer-to-Peer Netw. Appl. (2021) 14:888–897

in Section 4. As qw,j i will change according to qr,ij , an Here Cj i,1 is the ratio of the total request made by
q i
estimate of qw,j
r,ij
is desirable. the node to the download capacity of the node, i.e. how
Assuming that reputation of node i for node j is rj i , we much node over requested, and Cj i,2 is the ratio of total
can see that, capacity shared by all the nodes to the total requests made
in the network by all the nodes. Cj i,2 is statistically the
qw,j i = f (qr,ij , rj i , νj ).
average upload capacity against the unit request made by
Here, νj is an arbitrary constant chosen by node j and will a node in the network. Multiplying Cj i,2 with Cj i,1 will
range from 0 to 1. Normally, the value of νj will be equal to give us the average upload capacity per unit download
1. One possible way to decide qw,j i can be, capacity of the node. We have assumed that uncertainty has
identical distribution for every sample, and all the samples
qw,j i = qr,ij · (rj i )x · νj .
are independent, i.e. the noise samples are i.i.d.. Hence,
Here x is a reputation exponent. Nodes can decide we expect every sample to have the same variance. Let us
individually on the mechanism of computing qw,j i . assume this variance be σ . Estimate for qw,j i i.e. q̂w,j i is
Generally with higher rj i and qr,ij , a higher qw,j i should derived in Section 5.1.
be chosen. After deciding willing service rate for all the Once a node gets a offer from some node, it may act in
requesting nodes, two different conditions are possible, three ways, first, it may completely accept the offer, second,
 it may partially accepts the offer or third it may not accept
qw,j i ≤ Bj ,
the offer as mentioned in point 1 in Section 4. So
i
=⇒ ∀i, qo,j i = qw,j i . qo,j i = qij,ay + qij,an . (7)
or Here qij,ay and qij,an are accepted and not accepted data

qw,j i > Bj =⇒ qo,j i ≤ qw,j i rates respectively. The reputation must be given on the basis
i of qij,ay and not qo,j i as it is most likely the actual transfer
such that rate. It is given by,
  1−ηi
qo,j i = Bj . qa,j i q̂w,j i
i tij = × . (8)
min(qij,ay , qf,j i ) qr,j i
Here Bj is the shared bandwidth of node j and qo,j i is the
Here qa,j i is actual service rate i.e. the average rate at
offered service rate from node j to node i.Thus we can write
which receiver receives the data, qf,j i is feasible service rate
qo,j i [n] = qw,j i [n] − wj i [n] (point 2) i.e. the rate at which the TCP algorithm can get the
qo,j i [n] qw,j i [n] wj i [n] throughput via underlying link with packet loss probability
=⇒ = − (2) p. For the analysis and simulation purpose, we take TCP
qr,ij [n] qr,ij [n] qr,ij [n]
reno algorithm. With this algorithm, the feasible service rate
qo,j i [n]
Here n is time instant. qr,ij [n] is observable quantity and can be computed using the expression [35].
wj i [n]
is the uncertainty due to demand variation on node j
qr,ij [n] ⎛ ⎞
in the network. Equation 2 can be written as, ⎜ Wmax 1 ⎟
qf,j i ≈⎜
⎝ RT T ,   ⎟.
⎠
Qo,j i [n] = Qw,j i [n] − Wj i [n] (3) RT T 2bp
3
3bp
+ T0 · min 1, 3 8 p(1 + 32p )
2

qo,j i [n] qw,j i [n]

Here Qo,j i [n], Qw,j i [n] and Wj i [n] are qr,ij [n] , qr,ij [n] and (9)
wj i [n]
respectively. We can see that
qr,ij [n] Qw,j i [n] = (rj i )α
· νj
will remain constant over the time, so it can be replaced by Here qf,j i is the feasible service rate as a function of packet
Qw,j i . Let us assume that mean value of Wj i [n] is Wj i and loss probability p; Wmax is the maximum window opened
the ratio of Wj i and Qw,j i is Cj i given by by the receiver; RTT is the round trip time between the two
 nodes; T0 is the time-out period and b is the number of
1 − Cj i,1 ×C
1
, if Cj i,1 × Cj i,2 > 1. packets acknowledged by a single acknowledgement.
Cj i = j i,2 (4) 1−ηi
0, otherwise. qa,j i
Here the factor min(qij,ay ,qf,j i ) ensures that node
Here is not playing a game by offering more data rate and not
total datarate request made by the node serving it. Value of ηi will decide the amount of penalty in
Cj i,1 = (5) case of cheating by the offering node. This means that node
download capacity of the node
will be punished for offering more and actually providing
total capacity shared by the nodes in the network less. The ηi = 0 means maximum punishment whereas
Cj i,2 = . (6)
total capacity requested by the nodes in the network ηi = 1 means that no punishment is given to the node for the
Peer-to-Peer Netw. Appl. (2021) 14:888–897 893

difference in the offered and actual provisioning. The ηi can So

be decided by a node based on its experience with network.  
Wj i
To compute the value of ηi , node i can take the exponential S = 1− 11×M . (13)
qa,j i Qw,j i
moving average of the ratio min(qij,ay ,qf,j i ) , i.e.,
Here 11×M is a 1 × M matrix of 1s i.e. [111111.............1]T .
βi (n − 1) + ηi (n − 1)
ηi (n) = . (10) The covariance matrix
2 ⎡ 2 ⎤
 q i (n)  σ 0
Here βi (n) = ( j min(qij,aya,j(n),q )/ j 1 and ηi (1) = ⎢ 0 σ2 0 ⎥
f,j i (n)) ⎢ ⎥
βi (1). This value of ηi ensures that if the total download Cov = ⎢ . . ⎥ , (14)
⎣ .. .. ⎦
of the node is high than the penalty imposed should be
low i.e. if the node is already getting benefited from the 0 σ 2 M×M
network then it should not decrease the reputation of node
with more penalty. If node will not accept the offer, the Cov = σ 2 IM×M . (15)
qa,j i
value of qij,ay will be taken as a limiting case of 1 as both As we know that the BLUE [4] is
qa,j i = min(qij,ay , qf,j i ) = 0. Process of trust estimation
is explained in Fig. 2. St Cov−1 Qo,ji
Q̂w,j i = (16)
St Cov−1 S
5.1 Uncertainty estimation substituting the values,

Wj i
Based on observed values of trust, exact trust value can be 1 − Qw,j i
1t σ12 IM×M Qo,ji
estimated using some estimator. We have used Best Linear Q̂w,j i = Wj i

Wji
 (17)
Unbiased Estimator (BLUE) [4] for this purpose. Taking 1 − Qw,j i
1 t 1 I
σ 2 M×M 1 − Qw,ji 1
expectation in Eq. 3 solving, we get
E[Qo,j i [n]] = E[Qw,j i ] − E[Wj i [n]], 
n
= Qw,j i − Wj i . (11)
1
M Qo,j i [k]
k=n−M Qo,j i [n]
Q̂w,j i = Wj i
= Wj i
so the scaled mean will be 1− 1−
Qw,j i Qw,j i
E[Qo,j i [n]]
S[n] = (12) So, we can see that we need to compute the sample

mean of all the samples and the the ratio of noise mean and
where  = Qw,j i is the parameter to be estimated
parameter mean.
Qw,j i − Wj i Wj i This formulation have a problem that it is difficult to
S[n] = =1− .
Qw,j i Qw,j i compute the sample mean when the number of samples is

Fig. 2 Process diagram

894 Peer-to-Peer Netw. Appl. (2021) 14:888–897

600 iterations. For homogeneous network, it is same for every

Aq. P. = 0.1 node. Whereas, it is different in the heterogeneous network.
= 0.3 Requested node will transfer the data to the requesting node
Average absolute change in reputation

500
400
300
200
100
0 i. The N is total number of nodes.
0 100 200
Iteration
Fig. 3 Average absolute change in reputation for homogeneous
network
large. We can use moving average to estimate Qo,ij [n] as,
Qo,j i [n] = α · Qo,j i [n] + (1 − α) · Qo,j i [n − 1].
Here α depends on the rate of change of behaviour of nodes.
Initially Qo,j i [1] = Qo,j i [1].
Value of Cj i,2 is estimated regularly based on its
download capacity and total requests made. The value of
Cj i,2 is difficult to find as it is defined for the complete
network. So, we use the following heuristics to find it out.
We consider that nodes will gather the capacities shared and
requests made by their neighbouring nodes. Based on this
Old after the negotiation of resource transfer rate. At the end
of the slot, each node updates its reputation table as per
the quality of the transaction. First 50 iterations have been
taken as acquaintance period, i.e. a node will allocate their
bandwidth without referring to the reputation table.
We have plotted the average absolute change in
reputation estimation with increasing number of iterations
up to 500. Here, absolute change in reputation (t) is

t (itr) = |ti,j,itr − ti,j,itr−1 |. (19)
i,j
Here, ti,j,itr and ti,j,itr−1 are the reputation of node j after
iteration itr and itr − 1 respectively, as estimated by node
300 400 500 In Fig. 3, t is plotted by taking the average of
last ten measurements as proposed in [19] and according
to our proposed reputation estimation method for the
homogeneous network. In Fig. 4, t has been plotted for
the same methods but for heterogeneous networks. For the
above simulations, we have considered 200 nodes once with
α=0.1 and then with α=0.3. Homogeneous network means
(18) that all the nodes have the same download capacity and are
ready to provide resources to the same number of nodes.
Whereas by the heterogeneous network, we mean that nodes
have different download capacities and are ready to serve
different number nodes.
350
Aq. P.
= 0.1
= 0.3
300
Average absolute change in reputation

data nodes will evaluate the value of Cj i,2 locally and then Old
exchange this value with its neighbours. The values received
from different neighbours will be averaged to get a better 250
estimate of Cj i,2 . This process will be done periodically.
200

6 Numerical results
150
Performance of estimation method proposed in this paper
has been evaluated for 200 node network. We had 100
considered the discrete time instants for measurement and
estimation in the simulations. Every slot is termed as an
50
iteration. In each iteration, every node queries for some
resource, and after getting the reply, it requests the resource
from the replying nodes. A node asks for the resource, 0
from the nodes having the resource, as per its download 0 100 200 300 400 500
capacity. Requested nodes allocate their bandwidth as per Iteration
their reputation table in a probabilistic manner [36]. The Fig. 4 Average absolute change in reputation for heterogeneous
number of requests that a node will serve is fixed over all network
Peer-to-Peer Netw. Appl. (2021) 14:888–897 895

1 heterogeneous networks of 200 nodes for α = 0.1 and 0.3

respectively. It is evident from the figures that utilisation of
0.9 resources increases with the proposed trust estimation. This
again establishes the usefulness of the proposed method.
0.8

0.7
7 Conclusion
Utilization Level

0.6
= 0.1
= 0.3 In peer-to-peer networks, free-riding is a major problem that
0.5
Old can be overcome by using a reputation management system.
0.4 For any reputation management system, trust is a central
quantity. But, generally, it has been estimated through naive
0.3 methods in reported reputation management systems. These
methods do not consider certain important points. In this
0.2 paper, we have proposed an estimation technique using
0.1 BLUE. The proposed trust estimator considers uncertainties
Aq. P. in the trust estimation. The algorithm has also been tested
0 for a homogeneous and a heterogeneous network of nodes.
0 100 200 300 400 500 The numerical results show that the average absolute change
Iteration in trust values in the trust table of nodes is considerably
Fig. 5 Network utilisation by a homogeneous network of 200 nodes
less compared to any other older techniques. It implies that
reputation could be estimated more accurately using this
estimator. Better estimation of reputation also leads to a
This is evident in both Figs. 3 and 4 that using estimator better measure to counter the problem of free-riding in a
the change in reputation is less. This implies that by use peer to peer system.
of proposed estimator, reputation can be estimated more
accurately.
In Figs. 5 and 6, the utilisation level of shared References
resources in the network is plotted for homogeneous and
1. Osborne MJ (2009) Introduction to game theory: International
edition. Oxford University Press, Oxford. http://EconPapers.
repec.org/RePEc:oxp:obooks:9780195322484
0.9 2. Tang Y, Wang H, Dou W (2004) Trust based incentive
in p2p network. In: Proceedings of the e-commerce tech-
0.8 nology for dynamic e-business, IEEE International Confer-
ence, CEC-EAST ’04. IEEE Computer Society, Washing-
ton, DC, USA, pp 302–305. https://doi.org/10.1109/CEC-EAST.
0.7
2004.71
3. Adar E, Huberman BA (2000) Free riding on gnutella. First
0.6 Monday 5:2000
Utilization Level

4. Kay SM (1993) Fundamentals of statistical signal processing:

= 0.1
0.5 estimation theory. Prentice-Hall, Inc., Upper Saddle River
= 0.3 5. Buragohain C, Agrawal D, Suri S (2003) A game theoretic
Old framework for incentives in p2p systems. In: Proceedings of
0.4
the 3rd international conference on peer-to-peer computing, P2P
’03. IEEE Computer Society, Washington, p 48. http://dl.acm.org/
0.3 citation.cfm?id=942805.943825
6. Wang W, Li B (2003) To play or to control: a game-based
0.2 control-theoretic approach to peer-to-peer incentive engineering.
In: Proceedings of the 11th international conference on quality of
service, IWQoS’03. Springer, Berlin, pp 174–192. http://dl.acm.
0.1
org/citation.cfm?id=1784037.1784050
Aq. P. 7. Sherwood R, Lee S, Bhattacharjee B (2006) Cooperative peer
0 groups in nice. Comput Netw 50(4):523–544. https://doi.org/10.
0 100 200 300 400 500 1016/j.comnet.2005.07.012
Iteration 8. Lian Q, Peng Y, Yang M, Zhang Z, Dai Y, Li X (2008) Robust
incentives via multi-level tit-for-tat. Concurr Comput Pract Exper
Fig. 6 Network utilisation by a heterogeneous network of 200 nodes 20(2):167. https://doi.org/10.1002/cpe.v20:2
896
9. Dutta D, Goel A, Govindan R, Zhang H (2003) The design of a
distributed rating scheme for peer-to-peer systems. In: Workshop
on economics of peer-to-peer systems
10. Papaioannou TG, Stamoulis GD (2006) Reputation-based policies
that provide the right incentives in peer-to-peer environments.
Comput Netw 50(4):563. https://doi.org/10.1016/j.comnet.2005.
07.024
11. Marti S, Garcia-Molina H (2004) Limited reputation sharing in
p2p systems. In: Proceedings of the 5th ACM conference on
electronic commerce, EC ’04. ACM, New York, pp 91–101.
https://doi.org/10.1145/988772.988787
12. Kamvar SD, Schlosser MT, Garcia-Molina H (2003) The
eigentrust algorithm for reputation management in p2p net-
works. In: Proceedings of the 12th international conference on
world wide web, WWW ’03. ACM, New York, pp 640–651.
https://doi.org/10.1145/775152.775242
13. Xiong L, Liu L (2004) Peertrust: Supporting reputation-based trust
for peer-to-peer electronic communities. IEEE Trans Knowl Data
Eng 16(7):843. https://doi.org/10.1109/TKDE.2004.1318566
14. Zhou R, Hwang K (2007) Powertrust: A robust and scalable
reputation system for trusted peer-to-peer computing. IEEE Trans
Parallel Distrib Syst 18(4):460. https://doi.org/10.1109/TPDS.
2007.1021
15. Zhou R, Hwang K, Cai M (2008) Gossiptrust for fast reputa-
tion aggregation in peer-to-peer networks. IEEE Trans Knowl
Data Eng 20(9):1282. http://doi.ieeecomputersociety.org/10.1109/
TKDE.2008.48
16. Andrade N, Brasileiro F, Cirne W, Mowbray M (2004)
Discouraging free riding in a peer-to-peer cpu-sharing grid. In:
Proceedings of the 13th IEEE international symposium on high
performance distributed computing, HPDC ’04. IEEE Computer
Society, Washington, pp 129–137. https://doi.org/10.1109/HPDC.
2004.9
17. Yan Y, El-atawy A, Al-shaer E (2007) Ranking-based optimal
resource allocation in peer-to-peer networks. In: 26th annual
IEEE international conference on computer communications
(INFOCOM)
18. Xia Y, Song G, Zheng Y, Zhu M (2008) R2p: A peer-to-peer
transfer system based on role and reputation. In: Proceedings
of the first international workshop on knowledge discovery and
data mining, WKDD ’08. IEEE Computer Society, Washington,
pp 136–141. https://doi.org/10.1109/WKDD.2008.28
19. Satsiou A, Tassiulas L (2010) Reputation-based resource alloca-
tion in p2p systems of rational users. IEEE Trans Parallel Distrib
Syst 21(4):466. https://doi.org/10.1109/TPDS.2009.80
20. Huang MC (2019) A peer-to-peer reputation evaluation system.
In: Proceedings of 28th international conference, vol 64, pp 11–20
21. Meng X (2018) Suretrust: a super peer-aware trust model for
2-level p2p networks. Peer-to-Peer Netw Appl 11(3):619
22. Mengshu H, Xianliang L, Xu Z, Chuan Z (2005) A trust model of
p2p system based on confirmation theory. SIGOPS Oper Syst Rev
39(1):56. https://doi.org/10.1145/1044552.1044558
23. Liang Z, Shi W (2005) Pet: A personalized trust model with
reputation and risk evaluation for p2p resource sharing. In:
Proceedings of the 38th annual Hawaii international conference
on system sciences, HICSS ’05, vol 07. IEEE Computer Society,
Washington, p 201.2. https://doi.org/10.1109/HICSS.2005.493
24. Banerjee D, Saha S, Sen S, Dasgupta P (2005) Reciprocal
resource sharing in p2p environments. In: Proceedings of the
fourth international joint conference on autonomous agents and
multiagent systems, AAMAS ’05. ACM, New York, pp 853–859.
https://doi.org/10.1145/1082473.1082603
Peer-to-Peer Netw. Appl. (2021) 14:888–897
25. Song S, Hwang K, Zhou R, Kwok YK (2005) Trusted p2p
transactions with fuzzy reputation aggregation. IEEE Internet
Comput 9(6):24. https://doi.org/10.1109/MIC.2005.136
26. Selcuk AA, Uzun E, Pariente MR (2004) A reputation-based trust
management system for p2p networks. In: Proceedings of the 2004
IEEE international symposium on cluster computing and the grid,
CCGRID ’04. IEEE Computer Society, Washington, pp 251–258.
http://dl.acm.org/citation.cfm?id=1111683.1111799
27. Xu J, Du X, Cai W, Zhu C, Chen Y (2019) An online personalized
reputation estimation model for service-oriented systems. In: 2019
IEEE international conference on services computing (SCC).
IEEE, pp 116–120
28. Malik Z, Medjahed B, Rezgui A (2016) Scare: Reputation
estimation for uncertain web services. In: ACM Trans Internet
Technol, vol 16. https://doi.org/10.1145/2792979
29. Lee Y, Lee KM, Lee SH (2020) Blockchain-based reputation
management for custom manufacturing service in the peer-to-peer
networking environment. Peer-to-Peer Netw Appl 13(2):671
30. Zhao Y, Li Y, Mu Q, Yang B, Yu Y (2018) Secure pub-sub:
Blockchain-based fair payment with reputation for reliable cyber
physical systems. IEEE Access 6:12295
31. Zhuang Q, Liu Y, Chen L, Ai Z (2019) Proof of reputation:
A reputation-based consensus protocol for blockchain based
systems. In: Proceedings of the 2019 international electronics
communication conference, IECC ’19. Association for Computing
Machinery, New York, pp 131–138. https://doi.org/10.1145/
3343147.3343169
32. Awasthi SK, Singh Y (2020) Absolutetrust: Algorithm for
aggregation of trust in peer-to-peer networks. IEEE Trans Depend
Sec Comput: 1–1
33. Livan G, Caccioli F, Aste T (2017) Excess reciprocity distorts
reputation in online social networks. Sci Rep 7(1):1
34. Singha N, Singh YN, Gupta R (2020) Adaptive capacity
partitioning in cooperative computing to maximize received
resources. IEEE Access 8:3551. https://doi.org/10.1109/ACCESS.
2019.2926805
35. Padhye J, Firoiu V, Towsley D, Kurose J (1998) Modeling
tcp throughput: a simple model and its empirical validation.
SIGCOMM Comput Commun Rev 28(4):303. https://doi.org/10.
1145/285243.285291
36. Gupta R, Singha N, Singh YN (2016) Peer Peer Netw Appl
9(6):1101. https://doi.org/10.1007/s12083-015-0389-0
Publisher’s note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Ruchir Gupta received the
B Tech degree from HBTI
Kanpur and the PhD degree
from IIT Kanpur. He is
a faculty member at IIT
BHU. His research interests
include Applied Game The-
ory, Overlaid multi-casting,
Peer-to-peer networking and
Information diffusion in
distributed network.
Peer-to-Peer Netw. Appl. (2021) 14:888–897 897

Yatindra Nath Singh received Antriksh Goswami received

the BTech degree from REC the BTech degree from Uni-
Hamirpur and the MTech and versity of Rajasthan, MTech
PhD degrees from IIT Delhi. degree from IIITM, Gwalior
He is a professor in the Elec- and PhD degree from IIITDM,
trical Engineering Department Jabalpur. He is a faculty mem-
at IIT Kanpur. He had been ber at IIIT Vadodara. His
a chairman IEEE UP section. research interest include Dis-
He was given the AICTE Young tributed Computing, Applied
Teacher Award in 2002-2003. Game Theory, Computer Net-
He has supervised seven PhD works, and Peer-to-peer Net-
students. His research interests working.
include optical networkspro-
tection and restoration packet
and circuit switching, opti-
cal packet switching architec-
tures, overlaid multicasting, technology development for e-learning
and e-education, peer to peer networking, and optical communication
systems and theory. He is a senior member of the IEEE, a fellow of
IETE, and a senior member of ICEIT.