Você está na página 1de 8

Trust Models for Efficient Communication in

Mobile Cloud Computing


Florin Pop, Oana-Maria Citoteanu, Ciprian Dobre, Valentin Cristea, Mocanu Bogdan
Computer Science Department, University Politehnica of Bucharest, Romania
Emails: florin.pop@cs.pub.ro, citoteanu.oana@gmail.com, ciprian.dobre@cs.pub.ro, valentin.cristea@cs.pub.ro, mocanu.bogdan.costel@gm

Abstract—Aici pun abstractul meu The structure of the overlay network has a bio-inspired
Taking in consideration the large amount of data processed topology, a honeycomb. This model has been chosen because it
by distributed systems formed by servers and mobile devices, offers a fixed structure that can scale really well and support a
managing such large dimension of data has become a very
challenging issue with high impact on the end-user. Therefore, big number of users and it is auto-adaptive. Another objective
the management process of such systems can be achieved in an is to find an efficient way of forming the honeycomb structure
efficient way by using uniform overlay networks interconnected and to study how well it behaves when content is distributed
through secure and efficient routing protocols. The aim of this [19]. Another objective is to find a way of computing and
paper is to continue the research presented in the Trust Models storing the trust values of active users in a network, and to
for Efficient Communication in Mobile Cloud Computing paper
with new models of network overlays, written in UPB. see how these values evolve according to the behavior of the
de aici trebuie modificata idea users that joined a mobile network.
This paper describes a new model for resource trust manage- Mobile Cloud systems are characterized by availability,
ment in Mobile Cloud systems. It uses a bio-inspired structure massive scalability, robustness and flexibility. They are highly
for constructing the overlay in a honeycomb structure. Then adaptive, capable of self-organization and can offer load
we introduce a model for resource trust management based on
feedback, which is very similar to peer-to-peer trust management
balancing and fault tolerance. Some of their features are:
systems like EigenTrust, R2Trust, SuperTrust or TrustMe. Based efficient data search, redundant storage, hierarchical naming,
on overlay construction and trust value computation, we propose anonymity, trust and authentication. Because the nodes inside
a general flow for a resource request in Mobile Cloud systems a mobile network system can join the system and leave at any
logically organized as a honeycomb structure. We evaluate the moment without announcing their exit, the structure on which
proposed solution using simulation and the obtained results
highlights the performance of proposed auto-adaptive overlay
the system is build has to be able to respond in a quick and
that can be use as a content distribution platform in Mobile efficient manner to these changes.
Cloud systems. In the past years there has been a growing interest in
Index Terms—Resource Management; Overlay Networks; Ef- the scientific community for human-operated handheld de-
ficient Communication; Trust Management; Mobile Cloud; Peer- vices. Unfortunately, even though battery technology has been
to-Peer Systems
steadily improving, it has not been able to keep up with
the rapid growth of power consumption of mobile systems.
I. I NTRODUCTION
Mobile Cloud Computing (MCC) tries to extend the battery
The emerging number of mobile devices in the recent life by offloading the execution of power-hungry tasks onto
years and the highly developed computing infrastructure called clouds. An Auto-Adaptive Overlay Network for MCC, using
Cloud Computing has determined a new very interesting and opportunistic sharing of computing resources, can solve such
challenging domain in the field of research in Computer problems, by opportunistically spreading tasks in the network
Science called Mobile Cloud Computing (MCC). MCC bring of devices. Previous experiments on human mobility and social
together cloud computing with the mobility offered by the interaction, using Bluetooth, WiFi, ZigBee and other small-
mobile devices such phones, tablets, laptops and many others. and medium-range wireless communication protocols for co-
The most important challenges of MCC are concerning the location tracking, show the feasibility of establishing oppor-
performance through battery life of bandwidth, the environ- tunistic connections with mobile devices in close proximity
ment and security. Due to the fact that the number of users [22]. A first step in this direction has been taken at UPB.
for this kind of services is increasing exponentially new HYCCUPS is a preliminary platform developed at UPB to
business has arrived. The application markets offers the end evaluate the feasibility of resource sharing in case of co-
users services in Mobile Commerce, Mobile Learning, Mobile located smartphones [23]. It focuses on reducing the overall
Health-care and Mobile Gaming. consumed power of the interconnected devices. The main
A Mobile Peer-to-Peer Networks is a group of devices that differentiating factor from other platforms is that the offloading
communicate to each other in a decentralized way via wireless decision is taken considering both mobility and availability
communication technologies like IEEE 802.11, Bluetooth or of a node to offload a task, and not neglecting the overall
Ultra Wide Band. quality of experience of the users which are part of the cloud.
We primarily use the reduction of overall execution time
for mobile applications, reduction in power consumption, the
increase in robustness in case of failures, the enhance for user
experience, and uncertain sensor data as measure of progress.
The research in this paper is a new step towards optimization
of HYCCUPS’s running engine.
The paper has the following structure. Section 2 presents
the structure of auto-adaptive honeycomb overlay network by
presenting its role for Mobile Cloud, the network topology,
the routing algorithm and the overlay construction. Section 3
presents the resource trust management considering decentral-
ized approach and describes the trust solution for honeycomb
structure. The general flow for a resource request is also
described in Section 3. Section 4 presents the experimental
results that are split into two categories: overlay construction
and trust computation. Section 5 presents conclusion, future
work an a short case study about cases in which several users
are disconnected from the system, and the honeycomb ends up Fig. 1. Node coordinates
being split in different parts. Related work and similar projects
and examples are presented in all sections.
overview of four trust management schemes for Mobile Peer-
II. R ELATED W ORK to-Peer systems and further proposed a new reputation trust
In this section it will be presented a dig into the reputation management schema entitled M-trust.
trust management in decentralized Peer-to-Peer networks. Concerning the hybrid Peer-to-Peer networks, the authors
A good overview regarding reputation on Peer-to-Peer sys- of [7] suggested a Super-peer trust model called SuperTrust
tems has been created by the authors of [4]. In this paper it is that determines the peers to cooperate based on the common
presented a useful taxonomy of Peer-to-Peer reputation models interests. Therefore, peers that share less things in common
composed by gathering of information, ranking the peers and with others are likely to interact with others more frequently.
rewarding the peers. This model can handle with success several attacks like SM,
Having a good trust metric based on Social-Networks is a DM and CM attack.
major challenge because of the reputation ranking depends on III. SECTIUNEA 3
the peers social position. Therefore, the most dangerous peers
are the front ones. In paper [1], the authors proposed a better A. Routing Algorithm in Overlay Network
approach entitled Poisonwater for Social-Network based trust The honeycomb is formed inside a three-axis system: Ox,
metrics that is resistant to front peers attack than Eigentrust Oy and Oz. Each node is assigned a unique coordinate in
and Powertrust. In comparison the Poisonwater can mitigate the system. The coordinates change by following the edges
front peers attack by 20 parallel to the respective axis.
Another interesting approach regarding building good trust For example, if from the current node, defined by (u, v, w),
relationships in decentralized Peer-to-Peer systems is pre- we move on an axis parallel with Ox, the neighboring node
sented in paper [2], where the authors proposed new model of will have the same coordinate on Oy and Oz, and the Ox
trust based on Reputation and Risk evaluation. This model coordinate will be changed by 1, depending on the direction.
is suitable for defending against simple malicious attacks, If the vector formed by joining the initial node with the second
collusive attacks and strategic attacks. one has the same direction as Ox, then the x coordinate
In paper [3] is presented a new trust model that extends the will be incremented, giving the new position (u + 1, v, w).
D-S theory, that improves the filtering of false recommenda- If it is pointed in the opposite direction, the value must be
tions and the dynamic adaptive to strategic behaviors. decremented (u − 1, v, w).
Reputation management in Peer-to-Peer networks based on Figure 3 contains the coordinates of the nodes from the first
Distributed Hash Tables (DHT) is the main idea of paper [5]. 2 chains of the honeycomb structure.
The authors of this article proposed an algorithm that permits Let (u1 , v1 , w1 ) and (u2 , v2 , w2 ) be two nodes of the system
the peers in the network to get individual ranking values for and ∆x = u2 − u1 , ∆y = v2 − v1 and ∆z = w2 − w1 . Then,
each node based on knowledge exchange. the shortest path between the nodes will be given by |∆x|
Due to the high increase of the mobile devices in the edges parallel to Ox, |∆y| edges parallel to Oy and |∆y|
market, an important category of decentralized systems can edges parallel to Oz. When routing a message, the distance
be considered the Mobile Peer-to-Peer networks. Creating on each axis between the nodes must be shortened. First, the
trust and manage reputation in this kind of networks is a Ox coordinates will be compared, and, if the values differ, and
great challenge. Therefore the authors of [6] realized a brief there is a path that will shorten the distance, the message will
an edge if: |u2 − u1 | + |v2 − v1 | + |w2 − w1 | = 1. The
coordinates of the neighbors of any node can be obtained by
traveling across edges: (±1, 0, 0), (0, ±1, 0) and (0, 0, ±1).
Considering a node (u, v, w), we can compute a list of possible
coordinates for its neighbors: (u + 1, v, w), (u, v + 1, w),
(u, v, w + 1), (u − 1, v, w), (u, v − 1, w), (u, v, w − 1).
Out of these six triplets, only three define real neighbors.
The correct coordinates are the ones that respect the relation
1 ≤ u + v + w ≤ 2.
All the nodes of the honeycomb are divided into two
categories: white and black. They are placed so that any
edge will connect two nodes of different color. If a node is
white, then all its neighbors will be black. Considering a black
node, the vectors that connect it to his white neighbors are:
x+ = (1, 0, 0), y + = (0, 1, 0) and z + = (0, 0, 1). And the
Fig. 2. Route messages
ones connecting a white node to black ones: x− = (−1, 0, 0),
y − = (0, −1, 0) and z − = (0, 0, −1). If a white node has
the following coordinates (u, v, w), then its neighbors are:
be routed through there. If either of the previous conditions (u − 1, v, w), (u, v − 1, w) and (u, v, w − 1) (Eq.4). The
isn’t met, the Oy coordinate will be checked, and lastly, the neighbors of a black node are: (u + 1, v, w), (u, v + 1, w)
Oz coordinates. and (u, v, w + 1) (Eq.5).
Thus, a message sent between nodes A(4, −2, 0) and The nodes are connected as follows. The first chain of the
B(1, −1, 1) will pass through the following nodes, in this system is C0 . Each new node is added on the current chain, in
order: (4, −2, 0) → (3, −2, 0) → (3, −1, 0) → (2, −1, 0) → clock-wise direction until the chain is completed. The index
(2, −1, 1) → (1, −1, 1). of the first node on each chain is 0. The last value depends
A message from A(4, −2, 0) to C(3, 2, −3) will have the on the total number of nodes of the chain (see Eq.3).
following path: (4, −2, 0) → (3, −2, 0) → (3, −1, 0) → The first node of the chain is placed on the Ox axis. It
(3, −1, −1) → (3, 0, −1) → (3, 0, −2) → (3, 1, −2) → takes on the role of a SuperNode (SN) and is responsible with
(3, 1, −3) → (3, 2, −2). computing the node’s coordinates inside his chain.
The path sent between two nodes can differ depending on
IV. R ESOURCE T RUST M ANAGEMENT
the direction. For example, a message between C (3,2,-3) and
A (4,-2,0) will pass through the following nodes: (3, 2, −3) → Mobile Cloud systems are currently widely known and used
(3, 1, −3) → (4, 1, −3) → (4, 0, −3) → (4, 0, −2) → for sharing information: audio, video, images, binary files
(4, −1, −2) → (4, −1, −1) → (4, −2, −1) → (4, −2, 0). or simple text files using different Clouds. They also have
These paths are marked in Figure 4 with arrows. been used in e-commerce applications like: e-Bay, Amazon,
Yahoo!Auction or Edeal.
B. Overlay Network Construction According to [5], there is no universally accepted definition
Solution for overlay network construction is similar to for trust in computer science. In [6], it is defined as “The
approach presented in [3] for peer-to-peer networks. The belief the trusting agent has in the trusted agent’s willingness
authors describe a way of constructing a honeycomb overlay and capability to deliver a mutually agreed service in a given
by splitting the nodes of a chain into 2 different layers, context and in a given time slot”. Solution from [5] offers
called rings. Each node of the system is uniquely identified the following definition “the belief that an entity is capable
by computing a general index constructed from the values of of acting reliably, dependably, and securely in a particular
the chain, ring and index within each ring. The indexes from case”.
the interior chains are assigned so that they are identical to the In [6] we can also find the definition of reputation: “an
ones from the exterior ring of the previous chain. Each new aggregation of the recommendations from all of the third-party
node is added inside a chain in a clock-wise direction, until recommendations agents and their first, second and third hand
this is completed, and then the next chain is formed. opinions as well as the trustworthiness of the recommendation
A new construction method is proposed in this paper. It agent in giving correct recommendations to the trusting agent
makes use of the chains of the honeycomb, the indexes about the quality of the trusted agent”.
within each chain, which are set in a clockwise direction and The trust value is asymmetric, transitive and dynamic. It
the coordinates from the three-axis system. The connection is complex because there are many ways of computing its
algorithm is constructed by using the lemma given in [4]: value: as local or global, based on a single transaction or
nodes of a honeycomb of size k can be coded by integer triples on the history of the node’s interactions. It is asymmetric
(u, v, w), such that k+1 ≤ u, v, w ≤ k and 1 ≤ u+v+w ≤ 2. because the trusting node assigns its value to the trusted node
Two nodes (u1 , v1 , w1 ), and (u2 , v2 , w2 ) are connected by based on its previous behavior. This means that the value is
and not a centralized system. Each per is assigned an opaque
identifier so that it can remain anonymous. The status of
newcomers is lower than that of peers that have displayed
good behavior. This is needed so that malicious peers won’t
be encouraged to simply create a new identity when their trust
Fig. 3. Relationship model value becomes too low. A minimal overhead is introduced by
the messages passed inside the system. The system’s security
takes into consideration the existence of colluding peers.
computed for each node, it doesn’t describe the relationship There is two types of trust values that are computed: a
between the two nodes. The following scenario describes the local one and a global one. The former is computed based
transitivity property: if we have three nodes A, B and C for on the direct interaction a peer has with others. And the latter
which A trusts B and B trusts C, then A will trust C by a is based on the local values assigned to the peer by others.
value computed based on its trust for B and B’s trust for C. Each feedback is weighted based on the global reputation of
Trust is considered to be dynamic because its value is changed the peer that gave the value. The pseudo code of the algorithm
based on every transaction made, so that if a node starts to used to compute the reputation by EigenTrust [12] is described
send corrupted information, its trust value can be lowered. In below.
[7] the relationship between trust and reputation is described, The managers that compute and store the trust values use
and a computation model is presented in which the trust is a distributed hash table (DHT). The unique ID of the peer
mathematically computed based on the cooperative actions of (consisting of the IP address and port) is hashed in order to
a node towards another in a given context (see Figure 5). assign a peer its managers. The peers know the hash function,
Reputation is defined as perception formed through past so all the peers in the system know who the managers of a
actions about the intentions and norms of a node. And trust certain peer are. The algorithm for computing the global trust
as a subjective expectation that a node has towards another, value converges fast. Less than 10 exchanges of updated trust
based on the previous behavior and history. The flow of the values are needed.
system is the following: an exchange takes place based on the In order not to flood the trusted peers with requests, since
trust value. After it is completed, the reputation is updated and they would be chosen every time, the peers with which to
the trust value recomputed and so on. interact with are chosen based on probability. This is also
necessary in order to allow the newcomers of the system
A. Decentralized Systems to increase their trust values. The set of peers which have
This type of systems uses several users within the system downloaded files from peer i is Ai , and the ones from which
to store, compute and serve the trust values of other users. peer i has downloaded files is Bi .
Usually the users are required to give feedback after each • cij is the normalized local reputation value;
transaction. The algorithms for computing the actual trust • tik represents the trust that peer i places in peer k;
value are much more complex than in the systems previously • pi is a pre-trusted peer.
mentioned. Since the identity of the users that hold the trust Each peer i from the system has to execute the following
value is unknown, each system offers a way of initially as- steps:
signing these holders, and ways to route the messages through
the system, so that the values can be properly updated. Some 1: Query all peers j ∈ Ai for t0j = pj ;
examples of decentralized trust management systems are: 2: while δ <  do
(k+1) (k) (k)
EigenTrust [12], R2Trust [20], SuperTrust [13] and TrustMe 3: Compute ti = (1 − a)(c1i t1 + c2i t2 + . . . +
(k)
[21]. Overview. Usually, when a new user joins the system, cni tn ) + api ;
(k+1)
a few users are assigned with keeping track of his behavior. 4: Send cij ti to all peers j ∈ Bi ;
(k+1) (k)
One of the most important rules there are, is that the user itself 5: Compute δ = |ti − ti |;
cannot compute own trust value. Another is that there is no (k+1)
6: Wait for all peers j ∈ Ai to return cij ti ;
way of making sure that all of the users that respond to the trust 7: end while
value request are not malicious. This means that when a user
needs to interact with another, it is best to consider multiple The main objective of SuperTrust system is to protect the
responses, and to filter the ones that look like anomalies. security of peers and that of the trust values. The feedback
The EigenTrust algorithm [12] has been developed for a given is not decrypted when the peers rate the transactions, or
distributed file sharing system over peer-to-peer system. After when the results are aggregated [13]. The anonymity of the
each transaction takes place, it is rated so that the trust values peers remains hidden even in the case in which the feedback is
can be updated. A user is considered malicious, and his trust revealed. This is important so that the malicious peers cannot
value is decreased if it doesn’t send the requested data, or if launch attacks on the peers that reported low trust values for
the data it sends is corrupt or fake. them.
The system follows five design concepts. It is self-policing, Another important issue that is treated is that the given
which means the users themselves enforce the shared ethics, feedback cannot be modified or deleted.
For a feedback to be considered valid, the peer has to prove that won’t bring them direct value. This requires employing
that a transaction took place. And at most one feedback is a routing algorithm that offers the shortest path between 2
allowed for each transaction. In order to decrease the load on nodes, when routing the messages.
the Super Peers (SPs) and to tolerate more failures, the system The node that made the request for a certain resource rates
is implemented as a K-redundant network. Each peer has a set each interaction. It sends its feedback to the manager nodes
of n SNs that collect his feedback and aggregate the results (MP) of the one, which replied with the resource. The feedback
that other peers give about him. is given based on the time interval in which the request is
The general flow is presented below: completed, and on the validity of the data received. If the
1) Peer v sends a request for resource r to his SP; received data is corrupted in any way, the lowest possible
2) The local SP forwards the request to neighboring SPs; trust value is given. By considering the response time for
3) If a SP knows one of his peers has the resource, it the message in which a node announces it has the resource,
forwards this information to v along with the decrypted and the number of hops between the nodes, the trust value is
trust value of that peer; computed. If the distance is short, then the time interval in
4) After receiving a sufficient number of responses, v which the transaction is expected to take place is smaller. If
chooses the peer with the biggest trust value u; the distance is bigger, the time interval is longer. If the node
5) After the transaction ends, v encrypts his feedback about sends the needed resource in a shorter or equal time to the
u and sends it to his SP; minimum computed time interval, then the trust value is the
6) When the SP receives the feedback, it forwards the biggest allowed. As more time passes over the computed limit,
encrypted vote to its neighboring SPs; the trust value decreases.
7) The global trust value is updated without decrypting the Trust Value. A simple way of computing the trust value
intermediate reports. of a node is given in [14] where a solution for peer-to-
In order to be able to update the trust values, an encryption peer trust management is described. The new value is based
function that allows computations with encrypted values is on the feedback received for the last transaction, but it
used. All of the n SPs that are responsible for a peer share the also takes into account all the feedback values previously
same decryption key. In order to make sure a single feedback received: [T V new ] = [T V old ] + 1 (Eq.6) and {T V new } =
is considered for each transaction, when the peer submits its ([T V old ]{T V }old + F )/[T V new ] (Eq.7), where the [T V ]
vote, it attaches a time stamp to the message. All feedback are represents the number of feedback received for the node. The
considered invalid after a set period of time ∆t passes. The feedback value is F , and {T V new } is the new trust value.
messages are kept in a queue for that period of time, and when This way of computing the trust doesn’t account very well
the new trust value is computed, only the first is considered, for traitor nodes. After behaving correctly for a certain period
making it impossible to rate the same transaction twice. of time, they are able to make a large number of transactions
Applicability to Honeycomb Overlay Network. The fixed before their trust value is considerably lowered, as is shown in
structure of the honeycomb allows efficient message routing the experimental results section of this paper. A new model is
and assigning coordinates to nodes so that their anonymity can proposed, that doesn’t depend on the number of transactions
be protected. By routing the messages to direct neighbors, only that took place. It takes into consideration the current trust
these know the IPs of the nodes, which increases the system’s value of the node that is based on the previously received
security, and reduces the probability of attacks from malicious feedback, the current feedback value and the trust value of
nodes. the node that rated the transaction.
It also permits the choosing of managers, so that all the
nodes can compute their position within the system. This is
TYnew = αTYold + (1 − α)[TX F + (1 − TX )TYold ](Eq.8)
very important in order to reduce the number of messages
that are routed through the system, since the nodes can only where:
communicate only through direct neighbors. The managers can new
• TY is the new trust value of the node;
be protected from malicious nodes by assigning them so that old
• TY is the old trust value of the node;
they aren’t directly connected to the nodes whose trust values
• TX is the trust value of the node that gave the feedback;
they store and compute.
• F is the feedback value.
B. Proposed Model for the Honeycomb Overlay All values are bound in the interval [0, 1]. The feedback
When a new node joins the system, it receives a set of value is computed by taking into account the validity of the
coordinates. All the messages that are sent through the system data that has been sent and the time it took to fulfill the request.
are routed based on these coordinates. This increases the If the file is corrupt (the checksum is wrong), then the feedback
system’s security, since the only nodes who know the IP value will be 0. Otherwise, the value will be computed based
addresses of others are the direct neighbors. But it has a down on the time it took from the moment the initial request for the
point: nodes need to route messages that aren’t related to their resource has been made and the response that the node can
requests. The total number of routed messages needs to be kept share it has been received. The general flow for a resource
as little as possible so that nodes aren’t loaded with messages request is described below.
1) node A broadcasts a request for resource R;
2) All the available peers respond to A by sending their
coordinates;
3) For each response (ordered by response time), A will
send queries to 3 of the closest MNs to check the trust
value of the peers;
4) The MNs respond to the query with the trust value;
5) When A finds a peer B with an acceptable trust value, Fig. 4. Adaptive trust - false feedback
it makes a resource request;
6) After the transaction is finished, A sends the feedback
to the 3 managers of B;
7) The initial MNs send the feedback value to the others;
8) All MNs update their values.
By requesting the trust values only from the closest man-
agers, the values reach faster, and fewer messages are sent
through the system. It also prevents the feedbacks to be
routed through the peer whose trust value was requested. This
increases the security and restricts the attacks of malicious
peers. Fig. 5. Adaptive trust false feedback
All the messages are routed through the system by following
the shortest path between the peers. The exception is repre-
sented by the messages that are routed from the first 3 MNs to considerably lower than when connecting to a random node.
the other ones. These messages are routed so that the feedback When connecting the first 600 nodes (the 10th chain), the
never crosses through the peer about whom the feedback is maximum number for messages sent in the first case is 9,
given. while in the second are 31.
The trust managers are chosen when a peer joins the
B. Trust Management
system. Because the peers are assigned on each chain until
it is completed, this means that not all the managers can Both methods for the trust management are shown below.
be assigned from the beginning. As a new peer connects to Two main scenarios will be considered: when a node first
the system, it computes the peers from whom it should be connects to the system, and how a node behaves after a period
responsible, and queries the existent managers for the trust of time.
values. Adaptive Trust. Multiple values for α from (Eq.8) have been
In order to minimize the number of sent messages, the only considered. The first value considered is 7/8. This has been
broadcast message that can be sent is that for resource request. chosen because it is the default value for computing the RTT
To further optimize the system, the first request can be sent value for sending TCP packets [15].
only n hops through the system. If no answer is received, or The 7/8 value influences the trust values the least. When
if the trust values of the peers that have the resource are too a node first connects to the system, only after 7 transactions,
small, the distance is doubled. his value will be greater than 8. The 5/8 value requires only
2 transactions for the trust value to reach 8. For 3/4, it takes
V. E XPERIMENTAL R ESULTS 4. The values for the malicious nodes can be seen as being
The experimental results are split into two categories. First, symmetrical.
a way on how peers can connect to the system is shown, and The 3/4 value seems as the best choice for when a node first
then the adaptive method on how the trust can be computed joins the system. The correct nodes can gain a good trust value
will be presented. in a relatively short period of time, and the malicious ones are
detected early. The 5/8 value hasn’t been chosen because if a
A. Overlay Construction node receives a false feedback, then his trust value is lowered
For the construction of the overlay, two cases are taken too much. In order to gain a better perspective over how a
into consideration. When a new node wants to connect to false feedback can influence a node, the following case has
the system, it must first send the join message to the connect been considered. The trust values for the nodes that give the
node. This can be chosen from a random node that has already feedback are in the interval [0.87; 0.95]. The feedback values
connected to the system, or it can choose to connect to one range from [0.83; 0.95]. The initial trust of the node is 0.85.
of the SNs. The false feedback is given for the 3rd transaction and its
Figure 6 shows the number of messages that are sent within value is 0.3.
the system, in the worst-case scenario, until the new node finds The blue series in Figure 7 contains the trust values for the
its own position within the honeycomb system. The number real feedback. The green one is for the false feedback and the
of messages that are sent when choosing to connect to a SP is value α = 7/8 and the red one for α = 3/4. As it can be
seen, the trust value drops too much, from 0.85 to 0.74 in the [3] Bogdan Ghit, Florin Pop, and Valentin Cristea. 2011. Using Bio-inspired
case of the false feedback for α = 3/4. So, in order for the Models to Design Peer-to-Peer Overlays. In Proceedings of the 2011
International Conference on P2P, Parallel, Grid, Cloud and Internet
good nodes not to be easily influenced by malicious ones, the Computing (3PGCIC ’11). IEEE Computer Society, Washington, DC,
α = 7/8 value is has been chosen. USA, 248-252.
[4] Ivan Stojmenovic. 1997. Honeycomb Networks: Topological Properties
and Communication Algorithms. IEEE Trans. Parallel Distrib. Syst. 8, 10
VI. C ONCLUSION (October 1997), 1036-1042.
[5] Huaizhi Li; Singhal, M., Trust Management in Distributed Systems,
This paper describes a basic solution for auto-adaptive Computer , vol.40, no.2, pp.45,53, Feb. 2007
overlay network user for recourse trust management in Mobile [6] Mekouar, Loubna, Youssef Iraqi, and Raouf Boutaba. Reputation-based
trust management in peer-to-peer systems: taxonomy and anatomy, Hand-
Cloud Computing. It can be extended to offer better security book of Peer-to-Peer Networking. Springer US, 2010. 689-732.
mechanisms and to be fault tolerant. For the construction [7] Ding, Chen, Chen Yueguo, and Cheng Weiwei. 2009. A survey study
of the auto-adaptive honeycomb overlay network for Mobile on trust management in P2P systems. Department of Computer Science,
School of Computing-National University of Singapore. Technical Report.
Cloud users, two scenarios have been presented. Both have
[8] Ling Liu and Malcolm Munro. 2012. Systematic analysis of centralized
advantages and disadvantages. If a new node connects to the online reputation systems. Decis. Support Syst. 52, 2 (January 2012),
system by using a random node from the system, then the 438-449.
number of messages that are sent is significantly greater but [9] Silaghi, Gheorghe Cosmin, Alvaro E. Arenas, and Luis Moura Silva.
2007. Reputation-based trust management systems and their applicability
the security is better since everybody doesn’t know the real to grids. Core-GRID Tech. Rep.: TR-0064.
address of the Super Nodes. But, if a new Super Node were [10] Paul Resnick, Richard Zeckhauser (2002), Trust among strangers in
attacked, it would only have impact over how new nodes are internet transactions: Empirical analysis of eBay’ s reputation system, in
Michael R. Baye (ed.) The Economics of the Internet and E-commerce
added inside the system, and only for the last chosen Super (Adv. in App. Microec., Vol. 11), Emerald Group Publishing Limited,
Node. So, if the information about which node has been added pp.127-157.
to the system last is saved on other Super Nodes, then it [11] How do I leave Feedback? Description of E-bay’s feedback system,
http://pages.ebay.com/help/feedback/questions/leave.html.
would only delay the new connections until a new Super [12] Sepandar D. Kamvar, Mario T. Schlosser, and Hector Garcia-Molina.
Node can take the place of the one that has been attacked. 2003. The Eigentrust algorithm for reputation management in P2P
We can conclude that it is preferred for the new nodes to be networks. In Proceedings of the 12th international conference on World
Wide Web (WWW ‘03). ACM, New York, NY, USA, 640-651.
able to connect to the existent Super Nodes. Two methods [13] Tassos Dimitriou, Ghassan Karame, and Ioannis Christou. 2008. Su-
for computing the new trust values for nodes have been perTrust - a secure and efficient framework for handling trust in super
considered. The first one that takes into account the feedback peer networks. In Proceedings of the 9th international conference on
Distributed computing and networking (ICDCN’08), Shrisha Rao, Mainak
received for a transaction and the number of transactions and Chatterjee, Prasad Jayanti, C. Siva Ram Murthy, and SanjOy Kumar Saha
a second one that considers the feedback value and the trust (Eds.). Springer-Verlag, Berlin, Heidelberg, 350-362.
value of the node that gave it. For the latter, multiple values [14] Andreea Visan, Florin Pop, and Valentin Cristea. 2011. Decentralized
Trust Management in Peer-to-Peer Systems. In Proceedings of the 2011
for have been considered in order to manage the impact 10th International Symposium on Parallel and Distributed Computing
of the new feedback. Based on the experimental results, the (ISPDC ‘11). IEEE Computer Society, Washington, DC, USA, 232-239.
proposed model (Eq.8) with the value for =78 gives the best [15] Transmission control protocol, protocol specification
http://tools.ietf.org/html/rfc793.
results. The number of transactions doesn’t influence it, so the [16] Konstantinos Tserpes, Fotis Aisopos, Dimosthenis Kyriazis, Theodora
system is protected against traitor nodes. In the current form Varvarigou. 2012. A recommender mechanism for service selection in
of proposed solution, the malicious nodes can take advantage service-oriented environments. Future Gener. Comput. Syst. 28, 8 (Octo-
ber 2012), 1285-1294.
of the fact that when giving feedback; no proof for the fact [17] Talal H. Noor, Quan Z. Sheng, Sherali Zeadally, and Jian Yu. 2013. Trust
that the transaction took place is given. management of services in cloud environments: Obstacles and solutions.
ACM Comput. Surv. 46, 1, Article 12 (July 2013), 30 pages.
[18] Heiko Niedermayer, Ralph Holz, Marc-Oliver Pahl, and Georg Carle.
ACKNOWLEDGMENT 2009. On using home networks and cloud computing for a future internet
of things. In Proceedings of the Second Future internet conference on
The research presented in this paper is supported by Future internet (FIS’2009), Tanja Zseby, Reijo Savola, and Marco Pistore
projects: CyberWater grant of the Romanian National Author- (Eds.). Springer-Verlag, Berlin, Heidelberg, 70-80.
ity for Scientific Research, CNDI-UEFISCDI, project number [19] Dan Chen, Lizhe Wang, Xiaomin Wu, Jingying Chen, Samee U. Khan,
Joanna Koodziej, Mingwei Tian, Fang Huang, and Wangyang Liu. 2013.
47/2012 and SideSTEP - Scheduling Methods for Dynamic Hybrid modelling and simulation of huge crowd over a hierarchical Grid
Distributed Systems: a self-* approach, (PN-II-CT-RO-FR- architecture. Future Gener. Comput. Syst. 29, 5 (July 2013), 1309-1317.
2012-1-0084. [20] Chunqi Tian and Baijian Yang. 2011. R2Trust, a reputation and risk
based trust management framework for large-scale, fully decentralized
We would like to thank the reviewers for their time and overlay networks. Future Gener. Comput. Syst. 27, 8 (October 2011),
expertise, constructive comments and valuable insights. 1135-1141.
[21] Mustafa Al-Bakri, Manuel Atencia, and Marie-Christine Rousset. 2012.
TrustMe, I got what you mean!: a trust-based semantic P2P bookmarking
R EFERENCES system. In Proceedings of the 18th international conference on Knowledge
Engineering and Knowledge Management (EKAW’12), Annette Teije, Jo-
[1] Arduino project: http://arduino.cc/en/Guide/Introduction, and Raspberry hanna Vlker, Siegfried Handschuh, Heiner Stuckenschmidt, and Mathieu
Pi project: http://www.raspberrypi.org/about. d’Acquin (Eds.). Springer-Verlag, Berlin, Heidelberg, 442-445.
[2] Ted talk by Catarina Molta in which she encourages collaboration [22] A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott.
between different science branches Impact of human mobility on opportunistic forwarding algorithms. IEEE
http://www.ted.com/talks/catarina mota play with smart materials.html Transactions on Mobile Computing, 6(6):606 - 620. IEEE Press, 2007.
[23] [23] R.-C. Marin and C. Dobre. Reaching for the clouds: Contextually
enhancing smartphones for energy efficiency. In Proceedings of the
Second ACM Workshop on High Performance Mobile Opportunistic
Systems, HP-MOSys ’13, pages 31 - 38, New York, NY, USA, 2013.
ACM.

Você também pode gostar