An Incentive-Based Mechanism for Volunteer Computing using Blockchain
11An Incentive-Based Mechanism for Volunteer Computingusing Blockchain
ISMAEEL AL RIDHAWI,
Kuwait College of Science and Technology, Kuwait
MOAYAD ALOQAILY,
Al Ain University, UAE
YASER JARARWEH,
Jordan University of Science and Technology, JordanThe rise of fast communication media both at the core and at the edge has resulted in unprecedented numbers ofsophisticated and intelligent wireless IoT devices. Tactile Internet has enabled the interaction between humansand machines within their environment to achieve revolutionized solutions both on the move and in real-time.Many applications such as intelligent autonomous self-driving, smart agriculture and industrial solutions, andself-learning multimedia content filtering and sharing have become attainable through cooperative, distributedand decentralized systems, namely, volunteer computing. This article introduces a blockchain-enabled resourcesharing and service composition solution through volunteer computing. Device resource, computing, andintelligence capabilities are advertised in the environment to be made discoverable and available for sharingwith the aid of blockchain technology. Incentives in the form of on-demand service availability are givento resource and service providers to ensure fair and balanced cooperative resource usage. Blockchains areformed whenever a service request is initiated with the aid of fog and mobile edge computing (MEC) devicesto ensure secure communication and service delivery for the participants. Using both volunteer computingtechniques and tactile internet architectures, we devise a fast and reliable service provisioning frameworkthat relies on a reinforcement learning technique. Simulation results show that the proposed solution canachieve high reward distribution, increased number of blockchain formations, reduced delays, and balancedresource usage among participants, under the premise of high IoT device availability.CCS Concepts: •
Networks → Network management ; Network mobility .Additional Key Words and Phrases: Blockchain, Volunteer Computing, 5G, 6G, Internet of Things, AI.
ACM Reference Format:
Ismaeel Al Ridhawi, Moayad Aloqaily, and Yaser Jararweh. 2020. An Incentive-Based Mechanism for VolunteerComputing using Blockchain.
ACM Trans. Internet Technol.
1, 1, Article 1 (January 2020), 22 pages.
The future of intelligent and on-demand time-sensitive service provisioning relies heavily onsystem distribution and decentralization. At its early stages, smart city applications have relied oncentralized solutions such as the Cloud. Most applications (e.g. healthcare, autonomous drivingvehicles, etc.) have offloaded their tasks in terms of computing, storage and data analytics to clouddatacenters and storage sites [1]. IoT devices simply acted as data collectors with minimal datafiltration and analysis at their end. This was mainly due to the IoT devicesâĂŹ minimal hardware,software and intelligence capabilities. Moreover, cellular communication was restricted by low
Authors’ addresses: Ismaeel Al Ridhawi, [email protected], Kuwait College of Science and Technology, Kuwait;Moayad Aloqaily, [email protected], Al Ain University, UAE; Yaser Jararweh, [email protected], Jordan Universityof Science and Technology, Jordan.Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without feeprovided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice andthe full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored.Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requiresprior specific permission and/or a fee. Request permissions from [email protected].© 2020 Association for Computing Machinery.1533-5399/2020/1-ART1 $15.00https://doi.org/ ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. a r X i v : . [ c s . C Y ] S e p :2 I. Al Ridhawi et al. bandwidth availability, slow data rates, and minimal simultaneous device connections. Most devicesrelied on short distance communication such as wireless local area networks (WLANs) or wiredconnections. Such a solution seemed to be promising at first, given the low number of IoT devices.But with the enormous expansion in the number of IoT devices, in addition to the advancementsin user device capabilities in terms of hardware, software and communication, traditional cloudsolutions were no longer attractive, especially for time-sensitive applications such as autonomousself-driving vehicles [2], intelligent health monitoring [3], and emergency response services [4].At the early stages of research and development of the Fifth Generation (5G) communication net-work [5], but before its early stage deployment, alternative distributed and decentralized solutionswere made available to support time-sensitive applications, namely fog and mobile edge computing(MEC) [6]. Fog and MEC were introduced to provide processing and storage solutions located in thevicinity of mobile and edge IoT devices. This will somewhat eliminate latencies and communicationdelays experienced from the reliance on processing and storage cloud entities. Although suchsolutions were first experienced with cloudlets [7], fog and MEC have seen enhanced performancewith todayâĂŹs smart city and IoT ecosystems. To put things into perspective, fog computingand MEC provide an alternative central access point for not only communication purposes, butalso processing, storage, and intelligence. For instance, data that requires immediate attention andprocessing is handled by the fog, otherwise the job will be offloaded to the cloud. Alternatively,jobs submitted to the cloud can be offloaded to a number of fogs for processing or storage. In bothcases, this relieves the pressure off the cloud datacenter and ensures that all jobs are deliveredwithin the requested quality of service (QoS) and quality of experience (QoE) requirements [8].As technology has progressed at both the communication (e.g. 5G, MANETs, VANETs, etc.) andservice (e.g. fog and edge devices) layers, IoT and smart city applications started relying heavily ondecentralized and distributed solutions. As such, the concept of fog-to-cloud (F2C) communicationshifted more towards fog-to-fog (F2F) communication [8]. Both resource sharing and collaborationfor task completion became a necessity in order to complete jobs on time and meet the QoS andQoE requirements. Data replication and service availability at different fog sites made it possible foruser-specific services to be composed on demand [9]. Solutions for data and service decompositionwere developed to ensure that most of the data and simple services are available at fog sites,thus allowing for services to be composed [9]. Users requesting services that once were availableon the cloud, can now be composed and delivered in a timely manner. More complex services,especially ones that require machine learning techniques, still use resources of cloud datacentersand storage sites. Profit sharing mechanisms were also introduced to motivate cooperation andcollaboration among fog and MEC devices that belong to different internet and network serviceproviders (INP/NSP) [8].Although this provided an opportunity to overcome significant issues at earlier stages of thecloud distribution strategy, various user-specific requests (which arise as a result of new technologyavailability) still cannot be fulfilled at both the edge and cloud. For instance, multimedia user-specificservices that require the rendering of content (e.g. video and audio enhancements, color effects,and language support [10]), and which may not be available at the fog or cloud (or at least theadded rendering capabilities), can only be supported through end-device cooperation (i.e. resource,hardware and software capability sharing). A significant number of todayâĂŹs service requests,and most of tomorrowâĂŹs requests will require some type of artificial intelligence (AI) integrationto achieve enriched service capabilities. As such, reliance on the fog and cloud to fulfill all complexand composite service requests is no longer tolerable. Involvement of resource-rich IoT devicesin the service composition and delivery process is thus highly essential [11]. Decentralization,distribution, collaboration and cooperation, and resource and intelligence sharing at the end-devicelevel is required more than ever to not only achieve the requirements of most user- and device- ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:3 specific service requests, but also to achieve a balanced workload on all participants in todayâĂŹscomplicated networked ecosystem. Service requesters are no longer acquirers of consumableservices, but rather are involved heavily in the service provisioning process. As such the concept of volunteer computing in a tactile internet environment is highly needed for tomorrowâĂŹs beyond5G infrastructure, namely, 6G.Blockchain technology is being employed with several applications and integrated with othertechnologies such as IoT, health, energy, Fog/Edge, to name few [12][13][14]. The vision of 6G is toachieve total connectivity of intelligent things on-ground, in the sea, in the sky, and in space [15].To do so, we cannot simply rely on solutions that assume devices will cooperate and collaborate toshare their resources and capabilities in the service provisioning process.
Incentives in any formmust be given to participants to ensure fair usage and proper compensation for their involvement[16]. Such incentives may be in the form of service provider profit sharing or prioritized access tosubscribed services. As such, the significant majority of tasks will be conducted by collaborativeIoT devices with the aid of fog and MEC computing devices. Centralized entities such as clouddatacenters will act as a backbone to smart city applications and support the decentralizationprocess by offering intelligent processing and storage capabilities for very complex tasks thatcannot be delivered at the edge. Securely communicating data and collaborating to form and deliverservices can only be achievable with the aid of secure decentralized infrastructures that supportdevice to device communication. A plethora of applications will benefit from such incentivizedcooperative solutions. For instance, connected vehicles can share resources (e.g. computing, storage,power, etc.) as part of the cooperation process. In return, service providers get rewarded for suchon-demand requests. Multimedia content sharing is another hot topic in today’s social networking[6]. In densely crowded environments, such as stadiums, spectators can have on-demand access to agame replay with certain user-specific content enhancements from other spectators’ devices. Sucha collaborative environment will require some sort of an incentive model to ensure cooperationamong participants.This article proposes a cooperative and collaborative solution among edge IoT devices to sharetheir resources, computation, storage and intelligence capabilities. Blockchain is used as a form ofdecentralization to compose and deliver composite services securely and privately [17]. Incentivesare provided to participants to ensure that the developed framework is sustainable for all partici-pants, namely, both the service providers and requesters. The contributions of the proposed workare summarized as follows: • A cooperative IoT framework that supports volunteer computing at the end-user device levelto share their resources, processing, storage and intelligence capabilities. • An incentive-based mechanism is adopted to the framework to support and offer compensa-tion for participating in resource sharing and service composition processes. • A blockchain technique is adapted to support decentralized service composition and deliveryto sustain data privacy and consensus among participants.The rest of the article is organized as follows: Section 2 explores some of the most recent relatedwork in the literature. Section 3 formulates the problem and discusses the optimization aspect ofthe problem. The proposed IoT framework is considered in Section 4. Reward distribution andblockchain formation specifics are discussed in Section 5. Section 6 provides details in regards tothe conducted simulations. Finally, we conclude the article in Section 7 with some future workinsights.
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :4 I. Al Ridhawi et al.
The reliance on information and communication technology resources have grown exponentially,and data processing became the most important strategic resource. This is mainly due to theincrease in data volume. The solution to big data processing can be achieved through collaborationprocessing and data sharing, however, unwillingness to collaborate, the fear of resource sharing(i.e. trust issues), and sometimes inability to share due to connectivity problems are few examplesof persisting problems [18].Resource sharing techniques through collaboration (i.e. volunteer computing) are emerging asa mean to reduce repetitive tasks, faster computing, and promote the concept of decentralizedopen-source computing. However, we still face two main issues in regards to how to properlyincentivize participants to volunteer their resources and how to ensure the security and privacyfor the participants’ data. The work in [19][20] have proposed a mechanism of incentives throughadvancing the concept of data sharing based on blockchain and smart contracts. The former usessmart contracts to encourage users to share their data to overcome trust fear. Nash game equilibriumanalysis was used for incentives. Moreover, a reliable collaboration model for resource owners,miners, and trusted third parties have been proposed in [20]. The participant signs a smart contractfirst and then shares data and resources via blockchain. The incentive concept is being developedthrough revenue distribution among participants.Volunteer resources are not only limited to regular cloud participants but have been also in-vestigated at the vehicular network level. A privacy-preserving mechanism using blockchain forincentive announcement for communications between connected vehicles, namely, CreditCoin, hasbeen proposed in [21]. An aggregation protocol has been used for incentives. Similarly, the authorsin [22] proposed a new paradigm by merging vehicular ad-hoc network (VANET) with volunteercomputing to provide efficient utilization of idle computing resources. The authors evaluated theirproposed paradigm using job completion, throughput and latency to show better results comparedwith traditional approaches. Neither incentives nor blockchain were considered in their solution.When it comes to personal volunteer computing, in [23], the authors proposed a distributedcomputing approach that leverages personal devices (smartphones and laptops) to the personalcomputation needs from the general public of programmers to perform significant applicationsor community interest. Such a paradigm, also, encourages developers to maintain and enhancenew applications part-time, where no additional hardware is needed, and the process of tools wasdone over existing devices. While in [24], the authors have proposed volunteer computing as aservice (VCaaS) based edge computing, where volunteer computing resources are employed foredge computing to process data from IoT devices. Security and privacy were considered as well.To evaluate the effectiveness of such a new model, some researchers have studied the requirementsin addition to the strength of current volunteer computing platforms [25]. The authors have analysedmultiple issues such as the effectiveness of the active participants and how the computation andcommunication can be performed in addition to the analysis of task distribution and result validationpolices. On the other hand, the authors in [26] addressed the gap between the computational pillars.The authors drew on social psychology and online communities’ researches and proposed a three-dimensional model of the factors determining contributions of volunteer computing users (tenure,personal motivations and team affiliation). Also, the authors identified the relations between thesefactors and the actual contribution level.Among the many technical challenges emerging from this new technology, the most challengingproblem is task scheduling, where the resources are not only heterogeneous but also may go offlineat any moment. The authors in [27] proposed a deadline preference dispatch scheduling (DPDS)algorithm which is based on a dynamic task scheduling algorithm for heterogeneous volunteer
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:5 computing platforms. In DPDS, the task that has the minimum deadline constraint will completefirst by assigning it to a near volunteer node. Also, to maximize the number of computed tasksbefore the deadline constraint and to fully utilize volunteered resources, the authors proposed animproved dispatch constraint scheduling algorithm (IDCD) where tasks are selected according totheir priorities. The authors used a risk prediction model in the IDCD algorithm to ensure efficientapplication execution by predicting a completion risk of each task. In [28], the authors used a neuralnetwork mechanism to predict the job execution time and genetic algorithm, in order to distributejobs to volunteers with adjusting parameters to make it responsive to any changes. The resultsshowed the benefits of the proposed model even when volunteer computing network dimensionsare not high.As seen from the literature work, several challenges arise as a result of adopting this technologyat a large scale in terms of the heterogeneity of the resources, the variety of the capabilities, thedistribution of the tasks, the efficiency and utilization of volunteered resources. However, even ifall of these issues were managed and solved, retaining a large number of participants’ resources,encouraging data sharing, and guaranteeing continuous contributions are only possible if weprovide trust and proper incentives. This has not been explored yet, and we believe that this articlewill address those issues.
We consider a tactile internet network environment, as depicted in Figure 1, comprising of a plethoraof access points (APs) and base Stations (BSs) belonging to different ISPs and NSPs of differenttechnologies, such as LTE eNB, Wi-Fi APs, MEC servers, etc. Moreover, the network environmentcomprises a number of IoT end-devices
U E = { ue , ue , ... ue n } that have a set of resources andcapabilities defined as Cap = { cap , cap , ..., cap w } , and have been requested to cooperate in orderto complete a service request Req i = ( D i , Q i , O i ) . The request is defined through its descriptionproperties D i (defined later), the acceptable levels of specific QoS parameters Q i , and any otherrequirements O i such as cost or prioritized preferences. A service is composed of a set of tasks S j = { t , t , ... t m } , in which each task t has a size α t m , dependency on other tasks or sub-tasks β t m , and completion deadline γ t m . The complexity of a task will be measured by the participantin terms of its computation or storage intensity δ t m , defined in the form of processor cycles perdata block and energy consumption ζ t m . Such a measurement is not only dependent on the task’scharacteristics, but also varies in accordance to each node’s capabilities (e.g. hardware, software,etc.). Each participant ue n in the service composition and delivery process aims at maximizing itsparticipation gain G t n , m as defined in (1), by increasing its reward R t n , m as defined in (2), reducingits workload W t n , m as defined in (3), and eliminating/decreasing its penalties P t n , m as a result ofperforming task t m as defined in (4). Tasks with a large size α t m , more dependencies β t m , and strictercompletion time γ t m will lead to higher rewards R t n , m . Furthermore, achieving the requested QoSlevels for a task q t m within the set time limits will lead to higher reward values. On the contrary,not achieving the task (or achieving the task but not meeting the requested QoS and time limits)will lead to more penalties P t n , m , hence, less rewards. Penalties are incorporated within the gainfunction to ensure that participants are performing the requested tasks on time and in accordanceto the set service requirements. Nodes that simply join the composition process without adheringto the set rules (i.e. nodes participating in compositions beyond their resource capabilities for largereward returns) will receive penalties for not adhering to the set service requirements. This willensure that fair participation among cooperating nodes is achieved. Moreover, the workload W t n , m ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :6 I. Al Ridhawi et al.
Fog
IoT
DevicesMEC
WLAN
Cloud
IoT
Devices
Fig. 1. Tactile internet network environment comprising of a plethora of communication technologies, inaddition to volunteer computing. is dependent on the participants capabilities, such that an end-device that is described as energy-efficient with high and complex processing capabilities will complete a task with less workload(e.g. time, power usage, etc.). The requested/expected levels for reward, workload and penaltyspecifications, namely, χ t m = ( q t m , γ t m , δ t m , ζ t m ) is compared to the actual levels achieved at time t , namely, ` χ t m ( t ) = ( ` q t m ( t ) , ` γ t m ( t ) , ` δ t m ( t ) , ` ζ t m ( t )) . Having a solution where ` χ t m ≥ χ t m results inhigher rewards and less workload, which in essence leads to higher gains G t n , m ( χ t m , ` χ t m ( t )) . G t n , m ( χ t m , ` χ t m ( t )) = R t n , m ( χ t m , ` χ t m ( t )) − W t n , m ( χ t m , ` χ t m ( t )) − P t n , m ( χ t m , ` χ t m ( t )) (1) R t n , m ( χ t m , ` χ t m ( t )) = M (cid:213) m = t , q tm ≤ ` q tn , m ( t )≤ q tm τ q ` q t n , m ( t ) + M (cid:213) m = t , ` γ tn , m ( t )≤ γ tm τ γ max (cid:0) , ( γ t m − ` γ t n , m ( t )) (cid:1) (2) W t n , m ( χ t m , ` χ t m ( t )) = M (cid:213) m = t ` δ t n , m ( t ) + M (cid:213) m = t ` ζ t n , m ( t ) (3) P t n , m ( χ t m , ` χ t m ( t )) = M (cid:213) m = t , q tm ≥ ` q tn , m ( t )≥ q tm σ q ` q t n , m ( t ) + M (cid:213) m = t , ` γ tn , m ( t )≥ γ tm σ γ max (cid:0) , ( ` γ t n , m ( t ) − γ t m ) (cid:1) (4) τ q is the reward given for a participant that successfully completes the task within the qualitylimits set, namely, q t m ≤ ` q t n , m ( t ) ≤ q t m . The floor and ceiling QoS values are prone to changefrequently according to network performance. If the network resources are limited, then the offeredQoS value range for a particular service would be reduced to accommodate for the availableresources (i.e. participant resources). Details in regards to dynamic configurations of network ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:7 parameters is out of the scope of this article and has been discussed in [8]. τ γ is the reward pertime unit given for participants that successfully complete the assigned task at the deadline time γ t m , such that higher rewards are given for less time units ` γ t n , m ( t ) needed to complete the task. ` δ t n , m ( t ) is the processor/storage workload incurred on the participant for performing the giventask which resulted in the consumption of processor/storage resources at time t . Similarly, ` ζ t n , m ( t ) is the workload incurred on the participant for the power consumed to perform the given task. Theevaluation of the proposed systemâĂŹs energy consumption is measured in terms of the nodesworkload to complete a task. The computation intensity (i.e. CPU cycles per bit) to perform aservice task is considered when analyzing the energy consumption for a candidate participant. Weadopted the technique introduced in [29] to measure the power usage per CPU cycle. σ q is thepenalty incurred on the participant for performing the given task which resulted in QoS valuesbelow the requested levels. Similarly, σ γ is the penalty incurred on the participant for not meetingthe task deadline γ t m .As such, the objective that must be considered while distributing tasks among participants is tomaximize the gain achieved among all participants while adhering to obligations arising from therequested service as described in (5). P (cid:18) maximize N (cid:205) n = M (cid:205) m = t G t n , m ( χ t m , ` χ t m ( t )) (cid:19) s . t . C max N tm ⊂ N (cid:205) n = C n (5)The optimization problem is solved for collaboratively by all nodes whom are willing to participateand collaborate to deliver simple and composite services. Details in regards to the solution is lookedat in Section 5. The selection process is reliant on the rank given by other participating nodestowards the participant’s behaviour, which is dependent on previous successful task completionsand the node’s cooperation willingness characteristics. Hence, as seen in the constraint of (5), theset of participants which attain the maximum node cooperation and willingness score ( C n ), definedin Section 4, are selected for the composition and delivery process. The other issue that needs to be considered is the blockchain formation problem. Not only par-ticipants’ capabilities and their scores are considered, but also whether the result of performinga task by the participant is consistent with the input of the next block in the blockchain, hence,blockchain formation. The selection process must ensure that similarity measures between sequen-tial blocks are considered in the formation process. The goal is to formulate a composition path(i.e. blockchain) that increases the similarity score between one block and another, in addition tothe overall transaction for a service request. We consider the following optimization problem asdefined in (6). P (cid:18) max N tm ⊂ N (cid:205) n = M (cid:205) m = t Comp char ( ue n , t m , ue n + , t m + ) (cid:19) s . t . C max (cid:16) RW B comp (cid:17) (6)where RW B comp is the reward value for a service composition transaction using blockchain, attainedthrough a reinforcement learning algorithm. The reward value is determined as a result of previousrecords of blockchain formations’ experiences resulting from the selection of different blockchainpatterns. ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :8 I. Al Ridhawi et al.
Participants in the service sharing and composition process have two strategies, namely, participate or not-participate . The strategy is dependent on a number of factors: i ) the device’s capability (i.e.hardware, software, etc.), ii ) the user’s cooperation rationality given certain participation constraints, iii ) the device’s cooperation awareness in terms of continuous learning ability through analysis andstrategy readjustment. Upon joining a network environment as a participant in the service provisioning process, end-devices ( ue n ) communicate their capabilities Cap ue n set (see Example 1, presented in XML formatfor reader-friendly purposes) to the nearest MEC device either directly through point-to-pointcommunication or through other devices such as WiFi APs, device-to-device communication forAd Hoc networks, etc. The capability list and directory is consistently updated and shared amongall fog and cloud entities. Whenever a service request Req i is communicated to the participantswith a defined set of description properties D i , described through an ontological structure [30],participants compare the service request properties against their capabilities. Nodes can determinewhether the needed resources/capabilities are available using syntactic and semantic similaritycomparison against the request [31]. Details are out of the scope of this article and have beencovered in an earlier work [31]. Example 1 - Device capabilities for node 𝒖𝒆 𝒏 :
As described earlier, not only the intent for participants to maximize the gain from the cooperation isnecessary, but also to ensure that highly capable and cooperative nodes are joining the cooperationprocess to ensure diversified resource and service availability. A participant’s rationality towardscooperation is dependent on a different factors, namely, i ) the type and characteristics of the servicerequest, ii ) it’s current behaviour towards cooperative entities based on the current network statusand previous experiences, and iii ) the participant’s cooperative status. Other than the device’s capabilities towards achieving a service task, participant’s may decideto join or not join the cooperation process due to the type and characteristics of the service.Participants may have certain preferences in terms of what service tasks it may want to perform.For instance, a participant may decide to not join non-educational service requests or that ofanother characteristics. As such, upon joining a network environment, participants advertise theirservice characteristics participation preferences and priorities to ensure that they are excludedfrom non-preferred services. Example 2 provides an example of such advertisement.
Example 2 - Device service preferences for node 𝒖𝒆 𝒏 :
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :10 I. Al Ridhawi et al.
Comp char ( ue n , t m ) = (cid:12)(cid:12)(cid:12) char prefue n ∩ char t m (cid:12)(cid:12)(cid:12) w match (cid:12)(cid:12)(cid:12) char prefue n ∩ char t m (cid:12)(cid:12)(cid:12) + w match (cid:12)(cid:12)(cid:12) char prefue n ∪ char t m (cid:12)(cid:12)(cid:12) (7)where (cid:12)(cid:12)(cid:12) char prefue n ∩ char t m (cid:12)(cid:12)(cid:12) is the number of matching features when comparing the participant’stask preferences against the requested service task characteristics. Similarly, (cid:12)(cid:12)(cid:12) char prefue n ∪ char t m (cid:12)(cid:12)(cid:12) represents the non-matching features in the comparison. The participant’s behaviour towards other participants in the cooperative network is very crucial indetermining which set of devices will provide the optimal solution collaboratively. The behaviourof participants is dependent on different criteria, but most importantly the nature of the user.For instance, a participant that has had negative feedback as a result of cooperation with otherparticipants categorized under a certain characteristics category may pose strict conditions forcooperation for future service requests. Therefore, a ranking strategy is adopted to classify theparticipants’ behaviour. All participants involved in the cooperation process will rank each otherat the end of the service composition task, in addition to the serving MEC/Fog device and trustedentities, as depicted in Figure 2. This will ensure that a fair score is given to all participants andallows only those with an acceptable level of behaviour join the cooperation process and share thedistributed rewards. The figure outlines an example of where service requesters SR j , fog serviceproviders, and trusted entities T E x rank participants (whom volunteer their service capabilities Cap n ) in accordance to their behaviour, based on current and previous service composition processes.Ranking by requesters and trusted entities is only performed by those with direct cooperationtowards participants, whereas the fog ranks all participants. SR S ={t S1.1 ,t S1.2 }SR SR j S j ={t j.1 ,t j.2, , t j.m } . . . ue Cap ={t S2.2 } ue Cap ={t S1.2 ,t S2.3 } ue Cap ={t S1.1 } ue Cap ={t S2.1 } ue n Cap n ={t n } . . . S ={t S2.1 ,t S2.2 , t S2.3 } TE {ue ,ue } Fog TE x {ue n } . . . All Participants
SR = Service Requester, ue = participant device, TE = Trusted Entity, S = service, Cap = capabilities, t = task
Fig. 2. Requests in the form of service tasks are fulfilled using the capabilities of participant devices. Eachparticipant is ranked by the service requesters and fog/trusted entities that have knowledge about thebehaviour of participants, based on the current service composition process and/or any previous compositions.
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:11
We categorize participants’ behaviours using the following ranks/categories: • Highly Non-Cooperative - Participants of this category are described as devices that posesecurity threats to the network, regardless of whether the participant has been asked to joina composition task or not. Such participants need to be identified, reported, isolated andremoved from the network to avoid any future threats to the network structure. • Non-Cooperative - Such participants pose a security and/or privacy threat to the networkonly when requested to join a cooperative composition task. Hence, such devices need to beidentified, reported and isolated, but not removed from the network environment. Nodes ofthis type may request services but cannot participate in the service provisioning process. • Neutral - Nodes described as neutral pose no security issues to the network and may par-ticipate in the cooperation process. The results of whether the participant will sufficientlyattain the task requirements are unknown, and thus should only be requested to join thecomposition process when no other participant of a higher rank is available. Devices of thistype are usually casually interested in attaining rewards from cooperation tasks. • Partially Cooperative - Participants of this category have a fluctuating level of cooperativetask participation which is dependent on the type and characteristics of the service request,in addition to the network conditions. The objective of such participants is not only theamount of participation reward, but also the context of the service. • Cooperative - Such participants are considered cooperative at all times and the main objectiveof such nodes is to increase the reward value. The participant may sometimes decline theservice task request whenever the penalties and workload outweighs the rewards. Therefore,it is very important that rewards given for such participants are attractive enough to ensuresuch nodes join the cooperation process. • Highly Cooperative - This rank categorizes participants as highly cooperative in the sensethat in almost all circumstances, devices of this type will join the cooperation process evenif sometimes the task may lead to a loss in gain. Participants will join all cooperation tasksunless a security or privacy threat is posed by other participants.We adopt a weighted behaviour fuzzification function developed in [30] to give a score for eachcategory. This allows for the dynamic adjustment for the participants’ categories and providesaccurate and reliable participant scores. The fuzzified participant’s cooperative behaviour is definedin (8). C n = z cateдory (cid:32) N tm ⊂ N (cid:213) n = score t m ( ue n , ue n + ) (cid:33) (8)where score t m ( ue n , ue n + ) is the score in terms of cooperative characteristics that each participantend-device, fog and trusted entity provides to all other participants. z cateдory () is the fuzzificationfunction used to determine the fuzzified participant’s cooperative behaviour C n . Although incentives are provided to share resources and get involved in the composition process.Serving fog and MEC entities have authority to ban certain nodes not only according to theircooperation category, but also in accordance to their current and previous participation status. Forinstance, nodes that show greedy behaviour, where they only participate whenever highly valuablerewards are given may be banned from participation in future cooperation sessions to ensure fairdistribution of rewards to other novice participants. Moreover, a node itself may decide to whetherparticipate in cooperation/composition tasks or not. As such the participation status changes inaccordance to the participant’s desire, in addition to the context. We also note here that serving fog
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :12 I. Al Ridhawi et al. entities may ban nodes from participating in cooperation tasks if participants reject cooperationrequests excessively.
We assume that most participants have machine learning (ML) capabilities and at the same timecan achieve the task of federated learning collaboratively. Each participant adapts its own MLalgorithm and relies on the Stochastic Gradient Decent (SGD) method to perform stochasticapproximation of gradient descent optimization and replaces the data set gradient with an estimatedone through random sub-set data selection [32]. In terms of services that require distributed andcollaborated federated learning tasks, tasks’ outputs and trained models are either authenticatedusing the blockchain consensus algorithm or directed to the serving fog node for authenticationand aggregation. Such a decision on whether to use blockchain consensus or the fog is dependenton the type of service request. Services classified as ’ sensitive ’ are directed to the fog, otherwise theblockchain consensus method is used.Furthermore, with regards to whether to participate or not-participate in the service sharing andcomposition process, the training on data is performed locally on the end-user devices. Participantsuse previous gain achievements and other criteria in regards to cooperation of other participantsto determine whether future cooperative sessions are ideal in regards to the gains achieved. Localdata is also trained to avoid participating with other devices or share data that may be classified ashazardous leading to intrusion attacks [33].
Participants are selected in accordance to their advertised capabilities and the need for task distri-bution among a number of end-devices. As discussed earlier in Section 4.1, participants advertisetheir capabilities to the nearest fog/MEC device. From there, the information is shared amongneighbouring fog and MEC nodes. We classify the participant search process into two categories,namely, simple and complex . The former considers service compositions that rely entirely on the ad-vertised data by participants to fog/MEC nodes, where the participants are selected and a blockchainis formed to record the composition process, which is later used for the reinforcement learningprocess (discussed later). The latter, requires the aid of miners (i.e. trusted entities) to search forcapabilities not registered on the framework and requires coordination among participants andminers to complete the composition process on the blockchain. Moreover, the selection processamong the two methods is also dependent on the time-sensitivity and QoS restrictions.
Upon receipt of the service request from a requester for a simple service (i.e. one with restricted timeand QoS constraints or with availability of matching capabilities), a workflow plan is constructedto determine the tasks needed to be composed to deliver the composite service request. In additionto the tasks, the workflow plan identifies the best matching candidates in accordance to theircooperative characteristics scores as defined in (8). Figure 3 illustrates an example of a workflowplan constructed by the serving fog device. In our work, service tasks performed by participants inthe composition process are modelled using Workflow-nets which are an extension to Petri-nets[10]. Workflows guarantee the correctness of the cooperation and reachability problem [10]. APetri-net is a directed graph in which nodes are either transitions or places, where a place P isconnected to one or more transitions represented as Tasks . Transitions perform service tasks andare represented as tokens residing in places. A transition is said to be enabled only when there areno empty places (i.e. places with no tokens) connected to it as input. When a transition executes atask, tokens are removed from each of the transitionâĂŹs input places and tokens are created in
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:13 each of its output places. Moreover, workflows ensure that there is one place with no incomingtransition and one place with no outgoing transition. More details in regards to workflows arehighlighted in [10].
Service Task 2 to be provided by participant 𝒖𝒆 𝒏 + :
Task 3
Task 4 Task 5
Fig. 3. A workflow plan constructed by the serving Fog/MEC device outlining the set of tasks needed (describedas a transition) and the participants selected to perform the tasks (described as places) in accordance to theircooperative behaviour score as defined in (8).
Resource and capability acquisition requests are sent along with detailed information in regardsto the task description in terms of its size, complexity, dependency, etc... to the selected candidatesto identify their willingness to participate in the cooperation process. Each node calculates itsanticipated gain (if any) according to (1) and forwards the information to the serving fog node. Inorder for the participant to be selected and for the blockchain to be formed, the fog node will selecta set of candidates that achieve maximized gains according to (5). A blockchain is then formed fromthe participants to ensure that the composition process is guaranteed and recorded. The blockchainformation must consider (6), where the similarity score between two blocks (i.e. difference insemantic distance) is measured to ensure difference is minimized (i.e. similarity is maximized). Asmaller difference in semantic distance represents a beneficial move towards meeting the servicerequirements, indicating that the service request described through an ontology is nearly matchingthe outputs provided by the participants. The blockchain formation process adheres to positiveresults accumulated from previous similar blockchain formation trials. In essence, a reinforcementlearning process is adopted [8] to determine the reward value ( RW B comp ) defined in (9) that may beattained from similar blockchain formations and to speed up the formation process. RW B comp = N tm ⊂ N (cid:213) n = (cid:16) P (cid:0) ¯ Comp char ( ue n , ue n + ) (cid:1) × ˜ Comp char ( ue n , ue n + ) (cid:17) (9)where P (cid:0) ¯ Comp char ( ue n , ue n + ) (cid:1) is the probability of achieving the highest similarity for a selectedblock, ˜ Comp char ( ue n , ue n + ) is the expected highest similarity for a selected block. A matrix is ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :14 I. Al Ridhawi et al. formed for all the different blockchains that may be formed and the similarity achieved by eachselected block in the blockchain pattern. The value function for selecting a block from a set ofalternative blocks in a blockchain is therefore: V ue n ( t ) = V ue n ( t − ) + ρ (cid:0) V ue n − V ue n ( t − ) (cid:1) (10)where V ue n ( t − ) is the previous value function at time t − , ρ is the learning rate, and V ue n = RW B comp . As such, the blockchain which results in the highest value function is selected to ensurethat constraint C is met as defined in (5).Upon completion of the block selection process, all selected devices are informed and a blockchainis formed to complete the composition process and record all transactions on the blockchain. Therequested service is then delivered to the requester and rewards are distributed to all participants.All participants involved in the blockchain formation process then rank each other, in addition tothe serving fog node. The fuzzified participant cooperative behaviour score, defined in (8) is thenupdated. For service requests which require capabilities that are not registered among fog/MEC nodes, thesearch process is considered complex and requires the aid of miners to search for capabilities on theframework and coordinate with participants to complete the blockchain formation process. Sucha scenario can also be applied to cases with stringent QoS demands but relaxed time-sensitivityto ensure maximized QoS adherence. Upon formation of the workflow plan to determine theneeded tasks and capabilities, the process for participant selection with registered capabilities isidentical to that of a simple search process (described in Section 5.1). Participants with matchingcapabilities are selected as candidate nodes to be part of the blockchain. On the contrary, taskswith no matching registered/advertised capabilities will follow the complex search process. Miners(i.e. trusted entities) are notified of the capabilities needed to construct the block, which in essencewill be rewarded for their mining tasks. Miners must also ensure that participants’ cooperativecharacteristics adhere to the constraints defined in (5) and (6). Figure 4 depicts an overview of thecomplex search process.Miners are defined as fixed or mobile network and mobile devices that have the capability ofcommunicating directly with other end-devices through different communication methods (e.g.Ad Hoc). End-devices can gain the role of miners once the participant is labeled as trusted. Such alabel is given by fogs once the participant’s cooperative characteristics score exceeds a predefinedfog threshold, namely, C n ≥ ϑ . The threshold ϑ , is a dynamic value that changes in accordance tothe network condition. For instance, a network with few participants will have a relaxed thresholdvalue to ensure that the service composition and delivery process is achieved. On the contrary, ahighly dense network may have more stringent threshold values to ensure accurate service qualityadherence. We assume that the dynamic configuration process follows that of a tabu-search assistedvariable configuration optimization mechanism introduced in [8]. As such, end-devices havingboth roles, namely, participants and miners are capable of increasing there reward significantly.Miners collect rewards for participating in the search process. Tasks which require the aid of minerswill have the reward R t n , m ( χ t m , ` χ t m ( t )) shared among both the miners and selected participantend-devices. The portion of the share is dependent not only on the complexity needed to find theparticipant, but also finding other participants in the composition process that will provide accurateand stable blockchain formation which adheres to the overall QoS requirements. Therefore, thereward value for miners is determined as a portion φ of the reward value determined by the fog asdescribed in (11). ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:15 ue Cap ={t S2.2 } ue Cap ={t S1.2 ,t S2.3 } ue Cap ={t S1.1 } ue Cap ={t S2.1 } ue n Cap n ={t n } ... TE {ue ,ue } TE x {ue n } Participant
Devices TE {ue ,ue } TE {ue ,ue } ... P Task 1 P Task 2 Task m Miners ... Fog Workflow Plan
Tasks with no known registered capabilities are offloaded to the miners to search for candidate participants.
Fig. 4. Miners assist in the blockchain formation process by searching for end-devices with capabilitiesneeded to perform service tasks in cases of no registered capabilities or strict QoS requirements. R t T E , m = φ (cid:0) R t n , m ( χ t m , ` χ t m ( t )) (cid:1) (11)The blockchain formation process follows the goal of forming a composition path that reducesthe semantic distance (i.e. increases similarity) between the current output of the block and that ofboth the input of the next block and the service request. In essence, a blockchain is formed suchthat the result of the blockchain (i.e. composition process) increases the semantic similarity withthe service request. Figure 5 visualizes the blockchain formation process. From the figure, we seethat the output of the first block (i.e. service task performed by participant ue ) and the input ofthe next candidate block is compared to ensure that the one with the highest C n value is selected.At the same time, the semantic similarity between the output of the current block and the servicerequest requirement Req i is compared against that of the output of the next candidate block and therequirements Req i . Such a technique will guarantee that the strict QoS conditions of the requesterare delivered. Out B1 In B2 Out B2 In Bm B B B m Max(Comp char (Out B1 ,In B2 ))Max(Comp char (Out B1 ,Out B2 ))Max(Comp char (Out B1 ,In Bn )) Fig. 5. Selecting a set of blocks with the aid of miners to form a complete blockchain.
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :16 I. Al Ridhawi et al.
Additionally, before constructing the blockchain, miners report their candidate participants tothe fog, in which the former ensures that the participants in the blockchain achieve maximizedgains in accordance to (5). Algorithm 1 summarizes the process of forming a blockchain using thecomplex search method.
ALGORITHM 1:
Blockchain formation using the complex search procedure Input:
Service request
Req i is sent to the serving fog. If ( Req i has stringent QoS ∨ Cap missing requested capability cap w ) Construct workflow plan; For (each task t m with no cap w ) Request all available miners
T E to search for cap t m ; For (each
T E x ) Calculate
Sim ( T E x ) = Max ( Comp char ( Out B n , In B n + )) ; Calculate
Sim ( T E x ) = Max ( Comp char ( Out B n , Out B n + )) ; Calculate
Sim ( T E x ) = Max ( Comp char ( Out B n , In B m )) ; Send results
Sim ( T E x ) , Sim ( T E x ) , Sim ( T E x ) to fog; EndFor
Determine max N tm ⊂ N (cid:205) n = M (cid:205) m = t Comp char ( ue n , t m , ue n + , t m + ) ; Construct blockchain;
Distribute rewards to all participants according to R t T E , m and R t n , m ( χ t m , ` χ t m ( t )) ; EndFor
EndIf
Simulations were conducted using OMNET++ [34] and OverSim [35] as an overlay model forservice-specific overlays to mimic blockchains. Private blockchains are deployed with the SHA-256hash algorithm being used to ensure consistent and secure communication between participants. Infact, multiple private blockchains are created, one for each composition. The size of the simulatednetwork ares was × meters, with 10 MEC devices and up to 500 end-devices uniformlydistributed in the network. The number of trusted entities (i.e. miners) was set to 10% of the numberof participants. All end-devices, including miners act as both service requesters and providers.MEC devices act as 802.11g APs with a bandwidth of 54 Mbps, with both computing and storagecapabilities. All end-devices are mobile with a speed of 1-2 meters per second. Service task andcapability descriptions are specified in OWL/RDF format [36]. Capability characteristics similarityevaluations were conducted with the aid of OntoCAT [37]. The fuzzification and reasoning processeswere implemented using the jFuzzyLogic fuzzy engine [38]. Table 1 provides a summary of thesettings and configurations adapted in the simulator.The proposed incentive-based blockchain service composition technique with reliance on miners,referred to herein as Incentive-BC1 , is compared against i) the same solution without the useof miners, namely, with reliance on participant capability advertisements, referred to herein as Incentive-BC2 , ii) a non-incentive-based BC technique, referred to as non-Incentive-BC , and iii) atraditional fog-based service composition solution that does not rely on end-devices for servicetasks, referred to as non-BC . Evaluations were conducted in regards to resource usage, Blockchainformation hit ratio, the delay incurred for the blockchain formation process, and the total amountof rewards shared among participants in the service provisioning process. ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:17
Table 1. Simulator Settings
Simulation Parameters Numerical ValuesCommunication Protocol
IEEE 802.11g (for communication between UEs and APs)
Bandwidth
54 Mbps
Number of APs Number of UEs
Number of Miners
10% of UEs
UE Mobility Speed
Mobility model
Random Waypoint
Blockchain Hash Algorithm
SHA-256
Transmission/Idle Power
The amount of resource consumption was based on the average CPU usage per participant (eitherend-device or fog device) to compose and deliver the requested services. Results depicted in Figure6 show that by relying on the proposed
Incentive-B1 method, the CPU usage per participant sharplydrops by more than 70% when compared against the non-BC method. Such a result is very promisingand shows that MEC solutions can heavily rely on end-devices to perform service tasks and focusits responsibility on management rather than provisioning. Such a technique will also free upMEC devices to accept more service requests from clients. The use of trusted entities (i.e. miners)provides even further resource enhancements as shown in the figure when comparing the twosolutions, namely,
Incentive-BC1 and
Incentive-BC2 . Comparing the two solutions, a reduction ofnearly 8% in CPU resource usage is seen for service requests with 10 service tasks. C P U U s a g e ( % ) Service TasksIncentive-BC1Incentive-BC2non-Incentive-BCnon-BC
Fig. 6. A comparison of the overall CPU usage among four different methods, in terms of reliance on incentives,blockchain and service miners.
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :18 I. Al Ridhawi et al.
An evaluation of the energy consumption of the proposed scheme against the other methods wasconducted. The results depicted in Figure 7 are for a network density of 500 end-devices. Forthe non-BC solution, we assume that all service tasks are available at the fog devices. For theBC solutions, the service tasks are distributed among the end-devices, and hence require nodecooperation. Results show that the
Incentive-BC1 (with the aid of miners) and the
Incentive-BC2 bothprovide similar power consumptions, which outperform the non-incentive mechanisms. It shouldbe noted that although from the figure we see that less power is consumed for the
Incentive-BC1 technique in comparison to
Incentive-BC2 , this reduction is due to the offloaded tasks to miners toselect participants in the composition process. The overall reduction of the incentive mechanismsover the non-incentive mechanisms is an overall reduction of nearly 10% and 140% in powerusage when compared against the non-Incentive-BC and non-BC , respectively. The incentivizedBC solutions have shown that energy consumption at edge nodes is reduced dramatically and isshifted to the end-devices with less energy consumption at the end-device side. E n e r g y C o n s u m p t i o n ( J ) Service Tasks
Incentive-BC1
MinersIncentive-BC2non-Incentive-BCnon-BC
Fig. 7. A comparison of the overall energy consumption among four different methods, in terms of relianceon incentives, blockchain and service miners.
Testing the effectiveness of the proposed technique in terms of service composition success rate,namely, forming a successful and complete blockchain was considered in one of the experiments.The goal was to increase the number of service requests that arrive simultaneously at the fogdevices and observe whether the proposed solution, namely,
Incentive-BC1 can handle excessiveamounts of requests. As depicted in Figure 8, for the proposed incentive-based solution, the hitratio is nearly perfect for low to moderate simultaneous service requests. Moreover, for excessivenumbers of service requests, precisely with 100 simultaneous service requests, the
Incentive-BC1 solution outperforms all other techniques with nearly 80% success rate in blockchain formations.Additionally, we see that for the incentive based mechanisms (either with or without miners), the
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:19 blockchain formation hit ratio is nearly double that of non-incentive techniques, namely, non-Incentive-BC and non-BC . The hit ratios with 100 simultaneous service requests for
Incentive-BC2 , non-Incentive-BC and non-BC are 65%, 34% and 21%, respectively. B l o c k c h a i n F o r m a t i o n H i t R a t i o ( % ) Number of Service Requests
Incentive-BC1 Incentive-BC2 non-Incentive-BC non-BC
Fig. 8. Blockchain formation success rate in terms of the number of simultaneous service requests for fourdifferent solutions.
An experiment was also conducted to determine the delay encountered in forming blockchains (i.e.composing services) as the number of participants vary. A comparison of the proposed solutionagainst the three other techniques is shown in Figure 9. The figure shows the average experienceddelay, from the initiation of the service request and the completion of the blockchain formation,namely, composition of the requested service. The
Incentive-BC1 solution outperforms the othertechniques due to its capability of adapting to the time-constraints of the requested service. Nodeshaving time-sensitive delay requirements are carried out either at the fog site (if services areavailable) or carried out by the end-devices with the aid of fog devices and miners. With 500participants, the blockchain formation delay is reduced by nearly 19%, 49%, and 89% when comparing
Incentive-BC1 against
Incentive-BC2 , non-Incentive-BC and non-BC , respectively. We note here thatsome services cannot be composed (i.e. cannot form some blockchains) and hence are not consideredin the results. Results in regards to the ratio of non-successful blockchain formations are presentedearlier in Figure 9. Reward analysis was conducted on the proposed
Incentive-BC1 solution against the
Incentive-BC2 solution. The main objective of this test is to determine the proportion of rewards gained by minersagainst end-devices, and whether miners reduce the overall rewards distributed among end-devices.It was evident from the results, depicted in Figure 10, that the use of miners in
Incentive-BC1 solution resulted in the accumulation of more rewards for end-devices. This was evident from theincreased number of service requests being fulfilled given the aid of miners. For instance, with asimulation run of 500 participants, the total amount of rewards accumulated using the
Incentive-BC1
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :20 I. Al Ridhawi et al.
50 100 150 200 250 300 350 400 450 500 B l o c k c h a i n F o r m a t i o n D e l a y ( S e c ) Number of ParticipantsIncentive-BC1 Incentive-BC2 non-Incentive-BC non-BC
Fig. 9. Comparing the delay encountered in forming blockchains as the number of participants varies forfour different solutions. solution was 477 reward units (337 for end-device and 140 for miners). On the contrary, the totalamount of rewards accumulated using the
Incentive-BC2 solution was 292 reward units. That isan increase of 45 reward units just for the end-devices, in addition to the rewards distributed tothe miners. By comparing the proportion of rewards distributed among end-devices and minersusing the proposed
Incentive-BC1 solution, we see that miners have nearly a third of the rewards incomparison to end-devices.
350 50 100 150 200 250 300 350 400 450 500 R e w a r d U n i t s Number of ParticipantsEnd-Devices - Incentive-BC1
Miners - Incentive-BC1
End-Devices - Incentive-BC2
Fig. 10. Accumulated rewards for both end-devices and miners using the
Incentive-BC1 and
Incentive-BC2 solutions.
ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. n Incentive-Based Mechanism for Volunteer Computing using Blockchain 1:21
The vision of beyond 5G communication technologies is to provide connectivity for intelligentconnected things. End-devices will have the capability of performing sophisticated intelligent taskswith minimal or in most cases without reliance on centralized or semi-centralized entities likefog, MEC, and cloud computing. Such tactile internet infrastructures will enable seamless andsmart interaction between humans and machines to achieve revolutionized solutions for differentecosystems. This article introduced a cooperative IoT framework that relies on blockchain-enabledresource sharing and service composition through volunteer computing. Device capabilities areadvertised and made available for sharing using blockchain. Incentives in the form of rewardsare are given to participants to ensure fair and balanced cooperative resource usage. Minersare used to search for non-advertised service capabilities to ensure a fast and reliable serviceprovisioning framework. Experimental evaluations conducted in the form of simulations showedthat the proposed solution provides adequate and fair distributed rewards to all participants in theblockchain formation process. Moreover, high values of service hit ratio and balanced resourceusage among participants was also experienced under the premise of high IoT device availability.For future work, we plan to integrate the concept of federated learning with volunteer computing.IoT devices will collaborate together in the learning process and share their learnt models usingblockchains without reliance on any centralized training. This will ensure both data privacy andnetwork security.
REFERENCES [1] Mohammed Al-Khafajiy, Safa Otoum, Thar Baker, Muhammad Asim, Zakaria Maamar, Moayad Aloqaily, MJ Taylor,and Martin Randles. Intelligent control and security of fog resources in healthcare systems via a cognitive fog model.
ACM Transactions on Internet Technology .[2] Ala Abu Alkheir, Moayad Aloqaily, and Hussein T Mouftah. Connected and autonomous electric vehicles (caevs).
ITProfessional , 20(6):54–61, 2018.[3] W. Tang, K. Zhang, D. Zhang, J. Ren, Y. Zhang, and X. Shen. Fog-enabled smart health: Toward cooperative and securehealthcare service provision.
IEEE Communications Magazine , 57(5):42–48, 2019.[4] H. Bi, W. Shang, and Y. Chen. Cooperative and energy-efficient strategies in emergency navigation using edgecomputing.
IEEE Access , 8:54441–54455, 2020.[5] Moayad Aloqaily, Ouns Bouachir, Azzedine Boukerche, and Ismaeel Al Ridhawi. Design guidelines for blockchain-assisted 5g-uav networks.
IEEE Network , 2020.[6] M. Aloqaily, I. A. Ridhawi, H. B. Salameh, and Y. Jararweh. Data and service management in densely crowdedenvironments: Challenges, opportunities, and recent developments.
IEEE Communications Magazine , 57(4):81–87, 2019.[7] Ju Ren, Deyu Zhang, Shiwen He, Yaoxue Zhang, and Tao Li. A survey on end-edge-cloud orchestrated networkcomputing paradigms: Transparent computing, mobile edge computing, fog computing, and cloudlet.
ACM Comput.Surv. , 52(6), October 2019.[8] I. Al Ridhawi, Y. Kotb, M. Aloqaily, Y. Jararweh, and T. Baker. A profitable and energy-efficient cooperative fog solutionfor iot services.
IEEE Transactions on Industrial Informatics , 16(5):3578–3586, 2020.[9] Ismaeel Al Ridhawi, Moayad Aloqaily, Yehia Kotb, Yousif Al Ridhawi, and Yaser Jararweh. A collaborative mobile edgecomputing and user solution for service composition in 5g systems.
Trans. Emerg. Telecommun. Technol. , 29, 2018.[10] I. Al Ridhawi, Y. Kotb, and Y. Al Ridhawi. Workflow-net based service composition using mobile edge nodes.
IEEEAccess , 5:23719–23735, 2017.[11] A. Palade and S. Clarke. Collaborative agent communities for resilient service composition in mobile environments.
IEEE Transactions on Services Computing , pages 1–1, 2020.[12] Lewis Tseng, Xinyu Yao, Safa Otoum, Moayad Aloqaily, and Yaser Jararweh. Blockchain-based database in an iotenvironment: challenges, opportunities, and analysis.
Cluster Computing , pages 1–15, 2020.[13] Ouns Bouachir, Moayad Aloqaily, Lewis Tesng, and Azzedine Boukerche. Blockchain and fog computing for cyber-physical systems: Case of smart industry.
Computer , 53(9), 2020.[14] M. Aloqaily, A. Boukerche, O. Bouachir, F. Khalid, and S. Jangsher. An energy trade framework using smart contracts:Overview and challenges.
IEEE Network , 34(4):119–125, 2020.[15] Z. Zhang, Y. Xiao, Z. Ma, M. Xiao, Z. Ding, X. Lei, G. K. Karagiannidis, and P. Fan. 6g wireless networks: Vision,requirements, architecture, and key technologies.
IEEE Vehicular Technology Magazine , 14(3):28–41, 2019.ACM Trans. Internet Technol., Vol. 1, No. 1, Article 1. Publication date: January 2020. :22 I. Al Ridhawi et al. [16] C. Yang, J. Xiao, J. Li, X. Shao, A. Anpalagan, Q. Ni, and M. Guizani. Disco: Interference-aware distributed cooperationwith incentive mechanism for 5g heterogeneous ultra-dense networks.
IEEE Communications Magazine , 56(7):198–204,2018.[17] I. A. Ridhawi, M. Aloqaily, A. Boukerche, and Y. Jaraweh. A blockchain-based decentralized composition solution foriot services. In
ICC 2020 - 2020 IEEE International Conference on Communications (ICC) , pages 1–6, 2020.[18] M. J. M. Chowdhury, A. Colman, M. A. Kabir, J. Han, and P. Sarda. Blockchain as a notarization service for data sharingwith personal data store. In , pages1330–1335, 2018.[19] Shichang Xuan, Li Zheng, Ilyong Chung, Wei Wang, Dapeng Man, Xiaojiang Du, Wu Yang, and Mohsen Guizani. Anincentive mechanism for data sharing based on blockchain with smart contracts.
Computers & Electrical Engineering ,83:106587, 2020.[20] Meng Shen, Junxian Duan, Liehuang Zhu, Jie Zhang, Xiaojiang Du, and Mohsen Guizani. Blockchain-based incentivesfor secure and collaborative data sharing in multiple clouds.
IEEE Journal on Selected Areas in Communications ,38(6):1229–1241, 2020.[21] L. Li, J. Liu, L. Cheng, S. Qiu, W. Wang, X. Zhang, and Z. Zhang. Creditcoin: A privacy-preserving blockchain-basedincentive announcement network for communications of smart vehicles.
IEEE Transactions on Intelligent TransportationSystems , 19(7):2204–2220, 2018.[22] Ayesha Amjid, Usman Tariq, and Munam Ali Shah. Vanet-based volunteer computing (vbvc): A computationalparadigm for future autonomous vehicles.
IEEE Access , 2020.[23] Erick Lavoie and Laurie Hendren. Personal volunteer computing. In
Proceedings of the 16th ACM InternationalConference on Computing Frontiers , pages 240–246, 2019.[24] Tessema M Mengistu, Abdullah Albuali, Abdulrahman Alahmadi, and Dunren Che. Volunteer cloud as an edgecomputing enabler. In
International Conference on Edge Computing , pages 76–84. Springer, 2019.[25] Muhammad Nouman Durrani and Jawwad A Shamsi. Volunteer computing: requirements, challenges, and solutions.
Journal of Network and Computer Applications , 39:369–380, 2014.[26] Oded Nov, David Anderson, and Ofer Arazy. Volunteer computing: a model of the factors determining contributionto community-based scientific research. In
Proceedings of the 19th international conference on World wide web , pages741–750, 2010.[27] Ling Xu, Jianzhong Qiao, Shukuan Lin, and Wanting Zhang. Dynamic task scheduling algorithm with deadlineconstraint in heterogeneous volunteer computing platforms.
Future Internet , 11(6):121, 2019.[28] SS Parkhomenko and TM Ledeneva. Scheduling in volunteer computing networks, based on neural network predictionof the job execution time.
International Journal of Parallel, Emergent and Distributed Systems , 34(4):430–447, 2019.[29] S. Seng, X. Li, C. Luo, H. Ji, and H. Zhang. A d2d-assisted mec computation offloading in the blockchain-basedframework for udns. In
ICC 2019 - 2019 IEEE International Conference on Communications (ICC) , pages 1–6, 2019.[30] Y. Al Ridhawi and A. Karmouch. Qos-based composition of service specific overlay networks.
IEEE Transactions onComputers , 64(3):832–846, 2015.[31] Ismaeel Al Ridhawi, Safa Otoum, Moayad Aloqaily, Yaser Jararweh, and Thar Baker. Providing secure and reliablecommunication for next generation networks in smart cities.
Sustainable Cities and Society , 56:102080, 2020.[32] S. Shi, Q. Wang, K. Zhao, Z. Tang, Y. Wang, X. Huang, and X. Chu. A distributed synchronous sgd algorithm withglobal top-k sparsification for low bandwidth networks. In , pages 2238–2247, 2019.[33] Moayad Aloqaily, Safa Otoum, Ismaeel Al Ridhawi, and Yaser Jararweh. An intrusion detection system for connectedvehicles in smart cities.
Ad Hoc Networks , pages 87–88, 2009.[36] Grigoris Antoniou and Frank Van Harmelen. Web ontology language: Owl. In
International Handbooks on InformationSystems , pages 91–119. 2009.[37] Tomasz Adamusiak, Tony Burdett, Natalja Kurbatova, K Joeri van der Velde, Niran Abeygunawardena, DespoinaAntonakaki, Misha Kapushesky, Helen Parkinson, and Morris A Swertz. Ontocat–simple ontology search andintegration in java, r and rest/javascript.
BMC bioinformatics , 12(1):218, 2011.[38] Pablo Cingolani. jfuzzylogic: Open source fuzzy logic library and fcl language implementation.