Blockchain for Mobile Edge Computing: Consensus Mechanisms and Scalability
BBlockchain for Mobile Edge Computing:Consensus Mechanisms and Scalability
Jorge Pe˜na Queralta and Tomi Westerlund
Abstract
Mobile edge computing (MEC) and next-generation mobile networks areset to disrupt the way intelligent and autonomous systems are interconnected. Thiswill have an effect on a wide range of domains, from the Internet of Things to au-tonomous mobile robots. The integration of such a variety of MEC services in ainherently distributed architecture requires a robust system for managing hardwareresources, balancing the network load and securing the distributed applications.Blockchain technology has emerged a solution for managing MEC services, withconsensus protocols and data integrity checks that enable transparent and efficientdistributed decision-making. In addition to transparency, the benefits from a securitypoint of view are evident. Nonetheless, blockchain technology faces significant chal-lenges in terms of scalability. In this chapter, we review existing consensus protocolsand scalability techniques in both well-established and next-generation blockchainarchitectures. From this, we evaluate the most suitable solutions for managing MECservices and discuss the benefits and drawbacks of the available alternatives.
The scope of the Internet of Things (IoT) has been growing over the past decade, en-compassing an ever larger ecosystem that spans multiple domains. Some of the mostprominent research directions are smart cities [1, 2], vehicular technology [3, 4], orsmart healthcare systems [5, 6, 7]. In all these domains, a common factor is thatIoT systems are evolving towards more distributed architectures [8]. This shift from
Jorge Pe˜na QueraltaTurku Intelligent Embedded and Robotic Systems Lab, University of Turku, Turku, Finlande-mail: [email protected]
Tomi WesterlundTurku Intelligent Embedded and Robotic Systems Lab, University of Turku, Turku, Finlande-mail: [email protected] a r X i v : . [ c s . D C ] J un Jorge Pe˜na Queralta and Tomi Westerlund more traditional cloud-centric architectures has crystallized in the edge computingparadigm [9, 10, 11]. At the same time, novel technologies are increasingly designedwith decentralization in mind from their inception. Among these, blockchain tech-nology is set to be one of the key drivers behind the disruption of the technologicallandscape in the near future [12, 13]. Decentralized technologies are also the corner-stone behind the Internet 3.0 and Industry 4.0 revolutions that are undergoing [14].Blockchain technology is already a driver behind decentralized and distributedIoT systems, providing security [15], trust [16, 17], data management [18], peer-to-peer transactions [19], and fault-tolerand middlewares [20]. Blockchain platformscan be divided in two main types depending on how they manage user creden-tials, which have a direct impact on their applicability: (i) permissionless, or public,and (ii) permissioned, private, or consortium, blockchains. They differentiate in thatpublic blockchains are based on anonymous nodes with equivalent status, while con-sortium or private blockchains introduce different types of nodes and permissions,some of which require authentication in order perform certain actions. While trust inpermissionless blockchains is shared and distributed, in permissioned blockchainsthere is a series of validator nodes that represented trusted authorities [21].One of the main issues stopping a wider adoption of blockchain in IoT systemsis scalability, an inherent problem to Bitcoin’s architecture that multiple researchershave been addressing [22, 23]. While smart contracts have great potential in theIoT and distributed systems in general, their scalability and performance is closelytied to the overall performance of blockchain systems [24]. Nonetheless, multipleadvances in recent years have demonstrated that novel technologies can bring signif-icantly higher degrees of scalability and performance to next-generation blockchainsystems. Among these, Elastico provided the first implementation of a sharding pro-tocol in a permissionless blockchain [25]. Sharding is a technique that enables thedistribution of nodes in a blockchain into subchains for performing parallel vali-dation, thus increasing throughtput and reducing latency. A more recent scalableblockchain is OmniLedger [26], which reports better scalability than Elastico andpromises VISA-level latency and throughout if enough nodes form up the network.Owing to the distributed nature of blockchain systems, and distributed ledgertechnology (DLT) in general, IoT systems integrating them must already have a dis-tributed architecture by themselves. Therefore, it is only natural that blockchain isintegrated at the edge layer in most occasions, which represents the most distributedand interconnected layer of a typical IoT system. While sensors and actuators couldbe considered more distributed, they are not necessarily capable of node-to-nodecommunication. Through this chapter, we utilize the terms blockchain and dis-tributed ledger equivalently. However, distributed ledger technology (DLT) is oftenutilized to include more general systems that do not implement blockchains per se ,but instead rely on some other type of network or data management architecture. Anexample of this is IOTA, which utilizes acyclic directed graphs representing moregeneral data structures. The rest of this introduction delves into more details be-hind the nature of mobile edge computing and its integration with blockchain/DLTtechnology. lockchain for MEC: Consensus Mechanisms and Scalability 3
The European Telecommunications Standards Institute (ETSI) has promoted thestandardization of Multi-Access Edge Computing (MEC) [27], which shares theacronym with Mobile Edge Computing (MEC). The ”multi-access” term puts anemphasis on the multi-tenant infrastructure and better reflects non-cellular opera-tors [28, 27]. In this chapter, we do not make distinctions between the two termsas our focus lays on the role of blockchain with edge computing. MEC standard-ization has been led by the MEC Industry Specification Group (ISG) since the endof 2014. One of the main objectives of the ETSI MEC ISG is to define the basetechnologies for distributed and multi-tenant clouds that are meant to be deployedat the edge of the radio access network (RAN) [9]. By deploying data aggregationand processing tasks directly at the edge of the network, MEC services can providebetter reliability, lower latency and higher-throughput [29, 30, 7]. We will specifi-cally discuss throughout this paper how blockchain technology can play a key rolein terms of security and robustness for the resource management needed in a multi-tenant edge infrastructure, as well as enhance the services that MEC applicationscan provide [31, 32, 33, 34].One of the key architectural cornerstones enabling multi-tenancy and co-existingverticals at the MEC layer is network slicing [35]. Network slicing provides thebase for interfacing blockchain with other MEC services for a wide array of applica-tion scenarios [36]. Network slicing refers to the co-existence of multiple softwaredefined systems and networks (slices) sharing a common hardware infrastructure.Each of the slices can be thus designed independently and optimized for a particularapplication or business vertical [37]. In particular, slicing for vehicular communica-tion and offloading, together with 5G-and-beyond connectivity, are set to define themobility of the future [38].
The integration of blockchain within the MEC layer has been object of extensiveresearch over the past few years. Systems integrating blockchain and edge comput-ing can be roughly divided among those in which edge services are part of a largerblockchain system [39, 40, 41], and those in which blockchain is one of the servicesenhancing edge services [31, 42, 34, 43, 44, 33]. In this chapter, we are particu-larly interested in the latter type, as blockchain can provide a key piece in enablingtruly distributed, secure and efficient edge computing. With monetization of MECbeing a central topic of discussion since its early proposal [29], multiple works havefocused towards either enhancing security or utilizing blockchain as a marketplaceframework for users to access different applications at the edge [32, 34, 43]. Morerecently, other works have also delved into the potential of blockchain as a frame-work for managing edge resources [45, 46, 47, 44], as well as supporting autonomyin distributed robotic systems [36].
Jorge Pe˜na Queralta and Tomi Westerlund
From the security point of view, the integration of blockchain technology bringsevident benefits to edge computing. Among the main threats identified in a recentreport from the European Union Agency for Cybersecurity (ENISA) on 5G net-works and edge infrastructure [48], blockchain and DLT technologies can help ad-dress multiple remaining challenges. For instance, permissioned DLTs with built-inidentity management naturally provide an extra layer of resilience against mali-cious diversion of network traffic, manipulation of traffic, or authentication trafficspikes. When blockchain tehcnology is applied to resource management, it can serveas a framework to mitigate risks in terms of abuse of third party hosted networkfunctions, manipulation of the network resources orchestrator, or opportunistic andfraudulent usages of shared resources, among others. Moreover, safety-critical ap-plications can benefit from the enhanced security that blockchains and other DLTsprovide. These include the automotive sector with vehicle to everything communi-cation routed at the edge [49, 50], and the healthcare sector [34, 51, 52].
Multiple surveys and review papers have recently been published on the conver-gence of blockchain and mobile edge computing [53, 54, 55, 56]. Other surveysin either the blockchain or edge computing domains also mention the potential forintegrating one with another [57, 58, 59, 60]. In these and other works, scalabil-ity is often identified as one of the key aspects limiting the adoption of blockchainin edge computing. Nonetheless, these works describe the scalability problem ei-ther as a systemic blockchain problem [53], or from a system point of view [55].Most works also focus on a specific blockchain, Ethereum being the most widelyresearched blockchain for IoT [54]. In a blockchain, consensus algorithms are themain bottleneck in terms of scalability, i.e., the mechanisms enabling all nodes inthe blockchain network to validate transaction and stay synced. Depending on thetype of consensus algorithm, the scalability of the system might be limited by eitherthe computational complexity of the algorithm, or its communication complexity.We believe there is a gap in the literature describing how the consensus algorithmsaffect the scalability from these two points of view. Our objective is to bring furtherinsight in this area, providing a literature review and a discussion on the topic.In this chapter, we introduce the main consensus algorithms that form thebackbone of different blockchain solutions, including newer generation distributedledgers that do not follow many of the paradigms defined within the Bitcoin and suc-cessive blockchains. We then describe what can be the role of edge computing whenit integrates blockchain/DLT systems. In particular, we discuss the potential for thedifferent solutions in the IoT, from the point of view of scalability but also discussingthe different applications that are most suitable for different blockchain/DLT solu-tions. We do this from the point of view of consensus algorithms and their com-putational and communication complexity. Compared to previous works surveyingthe integration of blockchain and edge computing [53], we provide a novel classi- lockchain for MEC: Consensus Mechanisms and Scalability 5 fication of current research directions from an architectural point of view (Section3), while giving more insight into how the different consensus algorithms affect theintegration of blockchain/DLT and edge computing (Section 4).The rest of this chapter is organized as follows. In Section 2, we introduce themain consensus algorithms in blockchain systems and other DLTs, together withthe most prominent results in highly-scalable and low-latency blockchains. Section3 then reviews specific applications of blockchain at the MEC layer, and discusseshow the different consensus protocols integrate at the edge. In Section 4, we discusson the best blockchain/DLT solutions for different applications in the IoT, and hownext-generation systems that are currently under development might change the IoTand MEC landscape. Finally, Section 5 concludes this work.
In this section we start with the basics of blockchain technology and move into howthe field is evolving towards lower-latency, higher-throughput, and new conceptsaimed at increasing flexibility and scalability, such as sharding. We provide a his-torical point of view on the different consensus algorithms that have been proposedfor blockchains and other distributed ledgers, and include an overview of the mostprominent so-called third-generation blockchains.Consensus mechanisms are one of the key aspects within the design of decentral-ized networked systems or distributed computing systems. Consensus mechanismsare those algorithms that enable multiple independent agents to reach an agreementon a certain value, operation, transaction, or other types of data. In a distributed anddecentralized system, different agents, or nodes, need to be able to trust each other.Consensus mechanisms are the enablers of trust among agents. The most popularconsensus mechanisms to date in blockchain systems, according to a survey fromLi et al. [61], are proof of work (PoW), proof of stake (PoS) practical byzantinefault tolerance (PBFT) and delegated proof of stake (DPoS), with other significantapproaches including proof of authority (PoA), proof of elapsed time (PoET) orproof of bandwidth (PoB). Apart from some of the more traditional consensus algo-rithms listed above (e.g. PoW utilized in Bitcoin or Ethereum, and PoS being part ofEthereum 2.0 plans), in this document we also review consensus protocols utilizedin third- and fourth-generation distributed ledger systems such as the fast proba-bilistic consensus (FPC), and the cellular consensus (CC). We also put an emphasison defining the key technologies behind IOTA, a DLT designed for the IoT and anideal candidate for integrating DLTs with edge computing.
Jorge Pe˜na Queralta and Tomi Westerlund
Smart CitiesSmart HealthcareSmart Grids Agricultural IoTVehicular IoTUnmanned VehiclesApplicationsDistributed Ledger
Technologies
Ethereum and other PoW-based Blockchains
Hyperledger: Fabric, Sawtooth, Indy, Burrow, and Iroha
Third-Generation DLTs: IOTADistributedConsensusProtocols Proof of Work /Useful Proof of WorkPractical ByzantineFault ToleranceProof-of-Authority Proof of Stake / Delegated Proof of StakeFast Probabilistic ConsensusPaxos / Raft
Fig. 1: Blockchain/DLT Consensus protocols, systems, and applications in integra-tion with the Internet of Things.
Nakamoto’s proof of work designed for Bitcoin [62] has heavily influenced the de-velopment of new solutions for newer-generation blockchain systems. The PoW im-plementation in Bitcoin was a new application for an old algorithm. Originally pro-posed by [63] as a solution to deter spam activity from email senders, the main ideabehind PoW systems has remained unchanged: to request to all networked agentsto solve computationally intensive cryptographic problems in order to validate theiractivity, their identity, or those of another agent. In general terms, a PoW algorithmis, at its most fundamental level, an algorithm that solves a cryptographic problemwith a solution that is, in relative terms, hard to find and easy to validate. The com- lockchain for MEC: Consensus Mechanisms and Scalability 7 putational complexity of the validation of a PoW solution is therefore considerablysmaller than the complexity of finding such solution.Ethereum, the second most popular blockchain system after Bitcoin, also relieson PoW-based consensus to validate new blocks in the blockchain. A block canbe roughly defined as each of the entries in the distributed ledger that blockchainsimplement. A block does not include a single transaction, but often a set of trans-actions that are near in time. These transactions represent the block’s body, wheretransactions are defined in a generic manner and do not represent only the exchangeof cryptocurrencies. Transactions in PoW-based blockchains are not validated in-dividually, but instead all the transactions in a block get validated when the blockcontaining them is validated itself. A block is validated, or mined, by solving a PoWpuzzle. The original and most widely used puzzle in blockchains can be summarizedas follows: the PoW algorithm must find a block header, which is the result of apply-ing a cryptographic hash function to the content of the block body, satisfying somepredefined condition. However, for a fixed hash function and a fixed block body, theresulting hash will always be the same. In order to meet this condition (e.g., findinga hash smaller than a certain value), the algorithm must then find some other value,called a nonce, to be added to the current block body. Finding a nonce is the pro-cess often called mining. Once a block is mined, it is added to the blockchain and allother agents in the network can validate the solution. In Bitcoin and other blockchainsystems, the miner of a block gets a reward in the form of new cryptocurrency, thusmotivating nodes to participate in the transaction validation process.One of the problems of PoW-based blockchains is that two agents could solve aPoW puzzle at virtually the same time, for the same or different nonces. This cancreate two branches, or forks, in the blockchain. Nodes are situated in the branch ofthe solution that they received first. In Bitcoin, a built-in policy establishes that if onefork is longer than the other (or it accumulates more cryptographic complexity), thenall agents in the network judge it as the authentic one. This is a practical solution asit is highly improbable that two consecutive blocks will be solved simultaneouslyby two pairs of nodes. In any case, even if two or more blocks are solved at thesame time, at some point one of the forks will become longer. This defines the so-called 51% or double spending attack, as malicious nodes would need to to controlat least 51% of the network’s computing power in order to be able to introduce afaulty transaction in a block, validate it, and keep validating consecutive nodes inthe corresponding fork so that it is accepted as the canonical fork by the network.When the size of the network and the number of miners increases, the probability ofsuch attack is reduced, thus giving the blockchain its immutability and data integrityproperties.The benefit of having an expensive PoW solution in terms of hardware, energyconsumption and time is that it is equally expensive for malicious nodes to attackthe network. Part of the security of PoW thus comes from disincentivizing attackersbecause of the large a priori investment required in order to be able to attack and gaincontrol of the network, which would not pay off even if the attack is successful [64].
Jorge Pe˜na Queralta and Tomi Westerlund
Part of the research community has argued that taking into account the humongousamount of computational resources and electric energy put into mining to solvePoW puzzles, at least these could be defined in a way that the solutions found wouldhelp research in other fields. As an example, King et al. proposed the definition ofPoW puzzles that would find long chains of primes [65]. Solving these PoW wouldbe then dedicated to solve a mathematical problem which consists on finding thedistribution of the Cunningham prime chain. In this case, the Fermat Primality Testwould be used to validate the PoW solutions.A different research approach is the definition of simpler PoW requiring lesscomputational resources in order to reduce the entry barrier and provide a more uni-form distribution of mined currency. Pagh et al. introduced the concept of Cuckoohashing, in which the PoW difficulty would remain constant over time [66].
The basis for security and robustness in a PoW system comes from the amount ofcomputational resources needed in order to gain control over the network. Nonethe-less, this computational complexity also brings limitations. First, it limits the prob-ability for news nodes to be able to mine new cryptocurrency by themselves if theyjoin a large network. Second, it also limits the number of transactions that can be val-idated within a certain time interval. For instance, in Bitcoin, it takes an average timeof 10 minutes to validate a block and all the transactions it includes [67]. A differ-ent consensus approach that does not rely on computational complexity and that hasgained momentum in recent years is Proof of Stake (PoS). One of the main objectiveof PoS systems, which is being introduced, for instance, as part of Ethereum 2.0, isto reduce transaction validation latency. One of the first implementations of PoS in ablockchain system, which showed clear benefits in this direction, was demonstratedwith Nxtcoin [68, 69]. The idea behind PoS is to value the cryptocurrency that val-idating nodes put at stake , instead of their computational power. PoS mechanismselect validators with a probability proportional to the size of their stake, which isoften closely related to the amount of cryptocurrency that the node, or miner, owns.Nodes can lose the total value of their stake if they incur in fraudulent validations.In [70], a similar PoS system was proposed where the probability of selection of thenodes validating transactions was calculated based on both the pure stake and thestate of the block being validated in the blockchain.The 51% attack discussed in the PoW consensus mechanism is still a potentialattack vector in a PoS system. However, while in the PoW case attackers need toobtain control over 51% of the network’s computing power, which becomes in-creasingly easy as larger pools monopolizing the mining process are created, in aPoS system an attacker needs control over 51% of the cryptocurrency’s total supply.This is, in theory, a more difficult problem than gathering enough computing power. lockchain for MEC: Consensus Mechanisms and Scalability 9
Owing to the significant reduction of the computational complexity of the con-sensus algorithms with PoS when compared to PoW, the energy consumption foot-print is also reduced. PoS thus provides a more energy-friendly alternative whichin turn enables nodes with lower computational capabilities to participate in theblockchain as equals to all others. Multiple authors, such as [71] or [72], have stud-ied the sustainability of Bitcoin’s growth and its energy footprint, which researchersestimate to be the equivalent, on a yearly basis, to non-renewable energy resourcesconsumed by entire nations of the size of Czech Republic or Jordan. Nevertheless,this also means that because miners do not need to dedicate large amounts of compu-tational resources to mining, it is easier to perform Sybil attacks spawning multipleidentities within a single malicious node.In general terms, a PoS system relies on a validator or a set of validators which areeligible after depositing part of their stake. In other words, as described by Buterin et al. [73], nodes earn the right to propose a block only after locking part of thecoins they own on the blockchain. This is an extended definition over the pure PoSsystem firstly implemented in [74] as part of PPCoin, in which the total miner’sstake is directly considered.
The Practical Byzantine Fault Tolerance (PBFT) consensus algorithm was first pro-posed by Castro et al. in 1999 [75]. PBFT was the first algorithm with the ability tooperate in large asynchronous networks such as the Internet, while providing overone order of magnitude in processing power improvement over previous methods,allowing for high-performance Byzantine state machine replication, and demon-strating thousands of requests per second. Byzantine fault tolerance can be describedas the capacity of a system to maintain proper operation when multiple errors or un-expected behaviour occur within part of the system, but not its totality [76]. In adistributed network and considering the consensus problem, this is equivalent tothe ability of the network to provide a robust consensus even in an scenario wherea subset of nodes act maliciously, failing to forward valid data or sending invalidinformation.In a PBFT system, nodes are distinguished between validating and not-validatingpeers [77]. The validating nodes run the consensus algorithm, in which they replicatea state machine and evaluate its result. A client makes a request that is transmittedover the peer-to-peer network through the non-validating nodes, which act as prox-ies between clients and validators. Non-validating nodes do not participate in theconsensus mechanism, but are able to confirm the results. The PBFT algorithm isable to provide consensus across the network when at most one third of the nodesbehave arbitrarily or maliciously. Because the validator nodes need to arrive to thesame results regarding the client request, the state machine that is replicated mustbe deterministic.
In comparison with PoW and PoS systems, in PBFT individual transactions canbe confirmed without the need to wait for a block including several transactionsto be added to the blockchain. In terms of energy efficiency, PBFT requires lesscomputational resources than a PoW consensus, but increases the probability of aSybil attack, where a malicious node would create multiple instances pretendingto be a large number of parties. In practice, PBFT is often combined with a PoWthat must be solved in order to join the network and within certain time intervals toensure that every node in the network is dedicating some minimum computationalresources to the collective validation effort. An important benefit of PBFT over PoWand PoS is the low reward variance, as every node can be incentivized. This lowersthe reward variance across miners. Nonetheless, the scalability of PBFT is an issuedue to the large number of peer-to-peer communication exchanges required.
Excluding Bitcoin and Ethereum, which represent the majority of the cryptocur-rency market capitalization, one of the most successful blockchains within the IoTand industrial domains has been Hyperledger [78]. Launched in 2016 by the LinuxFoundation, the Hyperledger project is divided in five main subprojects whereblockchain frameworks for different aims are being developed: Fabric, Sawtooth,Indy, Burrow, and Iroha [79]. Among these, Hyperledger Fabric is the most popular,an enterprise-level and production-ready permissioned distributed ledger frameworkthat has already been applied across various industrial fields [80]. The aims be-hind the project include open-source and cross-industry development of an scalableframework for smart contracts. Through the rest of this chapter, we utilize Hyper-ledger to refer to Hyperledger Fabric unless otherwise specified.The consensus mechanism utilized in Hyperledger vary depending on the subpro-ject. For instance, Hyperledger Fabric relies on RAFT [81], while Hyperledger Indyutilizes Plenum, based on Redundant Byzantine Fault Tolerance (RBFT) [82]. Dif-ferent blockchains following the hyperledger design ideas rely on PBFT or adaptedBFT approaches.In recent years, blockchain technology has evolved towards a wider range of net-work definitions that do not keep the original structure of a blockchain in terms ofhow to store data within a distributed ledger. Among these, one of the most promi-nent distributed open ledgers under development is IOTA [83]. IOTA’s backbone isa directed acyclic graph that defines the tangle . The tangle is the underlying net-work upon which IOTA is built. While Bitcoin was born mainly as a distributedcryptocurrency, Ethereum evolved from it into a platform for smart contracts, andHyperledger is intended for industrial use, IOTA was specifically designed with theIoT in mind [84]. In IOTA, there are no miner or validator nodes confirming trans-actions, but instead each user must participate in the validation of two transactionsbefore being able to issue a new one on its own. This approach, together with the lockchain for MEC: Consensus Mechanisms and Scalability 11 tangle’s structure, makes IOTA highly scalable and free to use. IOTA’s developmentis open-source and led by the IOTA foundation.IOTA’s consensus protocol is defined within the Concordice system [85]. Themain differentiating aspect of IOTA’s tangle is the fact that multiple disconnectedsubnetworks can coexist for certain periods of time. This means, for instance, thatwhile a blockchain cannot contain two conflicting transactions in committed blocks,the tangle might temporarily contain two such transactions. IOTA deals with this,however, in a similar manner as Bitcoin does: the fact that a transaction is includedin the blockchain does not automatically mean it is valid, as two forks of the chainmight exist until one is deemed longer and this valid. Therefore, in both cases thereis only information about the probability of a transaction being valid, which in-creases as the blockchain, or the tangle, grow after that given transaction. In orderto make a decision on two conflicting transactions in IOTA and reach a consensusacross the network, Concordice proposes two consensus protocols: the fast proba-bilistic consensus (FPC) and the cellular consensus (CC). FPC, introduced in [86],is a leaderless probabilistic binary consensus protocol. FPC has low complexityfrom the communication point pf view, and is robust in a Byzantine infrastructure.As with PBFT, the basic idea behind FPC is voting. In any case, IOTA is still un-der development and is not production-ready. More detailed information on IOTA’sconsensus and CC is available in [87] and [88].Other DLT solutions claiming to be third-generation blockchain are Nano [89],with its underlying block lattice, and Skycoin [90], aimed at powering the Web 3.0.While Nano and IOTA are recent technologies, Skycoin has been under developmentfor several years and was born out of a series of external audits into Bitcoin, whichrevealed the different flaws in the PoW consensus protocol.
Second-generation blockchain systems, largely represented by the Ethereum blockchain,were defined as those introducing the ability of executing distributed programswithin the blockchain itself, therefore extending their applicability beyond cryp-tocurrency transactions and into the validation of more general types of transactions.These programs that can be executed within a blockchain are called smart contracts ,with one of the most notorious implementations being part of the Ethereum VirtualMachine and its corresponding stack [91], which provides a Turing complete lan-guage as part of its framework [92]. Ethereum also introduced a new programminglanguage to be dedicated to the development and implementation of smart contracts:Solidity [93]. Smart contracts as defined with Solidity code can be seen as a set ofinstructions defining transitions between states of the program, with both the datarepresenting the different states and the code defining the transitions being stored atspecific addresses within the Ethereum blockchain.In Ethereum, smart contracts are part of the Ethereum Virtual Machine (EVM) [94].The EVM is based on the existence of contract accounts in the blockchain, which ex- tend the functionality of external accounts, those controlled by a human or networknode through a public-private key pair. Contract accounts operate in an automatedway as a function of the code stored within the account. While external accounts aredefined based on their key pair, with an address determined based on the public keybeing assigned to each node joining the network, contract accounts have addressesthat are determined when the contract is created. In Ethereum, the address space isshared among both types of accounts. Contract accounts are created through transac-tions that have a null or empty recipient. Those transactions must contain code thatoutputs the smart contract’s code, which is then generated when the transaction’scode is executed within the EVM. In general terms, transactions including a payloadand Ether (Ethereum’s cryptocurrency) between external accounts in Ethereum areextended so that when a transaction’s target account is a contract account containinga set of code instructions, these are executed given the payload in the transaction. Akey concept in Ethereum is gas. Upon creation, transactions are assigned a definitequantity of gas. The gas is a measure of the processing power that will be dedicatedto that transaction. In other words, the gas is the transaction fee. The gas is initiallycharged into the transaction, and its reserve gradually decreases as a function ofa set of predefined rules when the EVM executes the different transaction instruc-tions. The gas that is left is refunded to the transaction creator. The gas price, whichis paid upfront, is decided by the creator node. Miners, which obtain the gas price asa reward, decide which transactions to mine based on the amount of gas included.Therefore, the gas price is decided based on the market and the desired priority fora specific transaction.
While second-generation blockchains introduced new functionality and improve-ments over Bitcoin-based blockchains at different levels, one of the main challengesin blockchain systems remained: scalability [95]. This is mostly due to the large andincreasing amount of computational resources required for mining. From the com-munication point of view, Bitcoin and other similar blockchains only require onebroadcast per block, and therefore the main bottleneck comes from computation(which cannot be directly decreased while maintaining security). In PBFT-basedsystems, multicast messages are required for validation, and thus the main scala-bility problem is the communication cost [22] (which cannot be directly reducedeither without compromising security and robustness of the consensus mechanism).Multiple research efforts have been directed towards the realization of more scalablesystems, with new blockchains based on PoS and PBFT showing promising results.Elastico, introduced in [25], was one of the first scalable blockchains that introducedthe concept of sharding: to divide the network in subnetworks, or shards , that wouldvalidate transactions in parallel. Elastico was the first blockchain system to providea full implementation of a sharding scheme for a permisionless blockchain. A differ- lockchain for MEC: Consensus Mechanisms and Scalability 13 ent early sharding proposal was presented in [96], where Merklix trees are utilizedto merge the state of the different shards into the global blockchain state [97, 98].Another blockchain system aimed at scalability that has had an important im-pact on subsequent research is OmniLedger [26]. Omniledger scales linearly withthe number of nodes in the blockchain, and reports transaction times able to matchcredit card standards if the size of the network arrives to a certain threshold. The keydifference with Elastico in terms of scalability is that in Elastico the network perfor-mance scales with the computational power in a linear fashion, while in Omniledgerit does so with the number of validator nodes. In Hyperledger, the scalability of thenetwork has seen significant improvements since the release of Fabric 1.1.0 [99].Moreover, the number of channels can be scaled with little to no impact on perfor-mance according to the same report.Perhaps the biggest effort that is currently being put into the development ofa truly decentralized, permissionless and scalable yet secure blockchain is the de-sign and development of Ethereum 2.0 [100], where huge amounts of computingresources will be no longer required for mining [101]. The Ethereum Foundationand other developers behind Ethereum 2.0 have embraced Proof of Stake as themain consensus mechanism, while still utilizing PoW to secure the network, andthe concept of sharding towards scalability. The consensus is based on the Casperprotocol [102], which incentives for mining have been described in [73]. The im-pact that shards have on transaction scalability is relatively clear, with a much largerthroughput being possible in terms of transactions validated per second. Nonethe-less, it is not straightforward to extend the implementation of smart contracts withsharding. As smart contracts have associated a series of data states corresponding totheir code, each state change can be though of as a transaction. Contracts can be ex-ecuted within a single shard, or a cross-shard synchronization mechanism must existto allow for data to flow between shards. In [26], the authors introduced introducedAtomix, a client-driven lock/unlock protocol, to ensure that a single transaction canbe committed across multiple shards, while enabling the possibility of unlocking re-jected transaction proofs in specific shards. The original Atomix state machine canbe extended to accommodate the execution of smart contracts across shards.
This section reviews and classifies the existing research in the integration of blockchainand MEC from an architectural point of view. We classify the different approacheson three main categories, illustrated in Fig. 2. The first category encompasses worksproviding a system-level integration where a blockchain is one of the key piecesat the heart of the edge infrastructure, managing services and resources. The sec-ond category includes approaches that utilize blockchain as a middleware betweenthe edge infrastructure (hardware and software) and the third-party services beingprovided through MEC. Finally, the last category comprises those works where the blockchain is part of individual applications, for aspects such as security or identitymanagement.In general terms, Ethereum is the most widely applied blockchain platform inthe IoT, owing to the maturity of its smart contracts framework enabling complexinteractions between data producers and consumers [103, 104]. In the same area,Hyperledger has potential to disrupt the IoT with more scalable solutions and theability to run distributed programs as chaincode [105]. In all these cases, nonethe-less, the blockchain runs in embedded edge gateways providing stable connectivity,and where enough power and computational resources is available. With the poten-tial to reach embedded devices at the sensor layer, and being developed specificallyfor the IoT, IOTA is set to play an increasingly important role. Owing to its lowinherent computational requirements and being highly scalable, IOTA is the idealcandidate for edge computing systems and hardware.
One of the most critical points at the edge is resource orchestration [29]. In order toenable a wide variety of use cases, multi-tenant applications, and ad-hoc deploymentof different modules, MEC infrastructure needs to be able to manage its resourcesin real time, while also orchestrating how the network is being utilized. This in-cludes processes from allocating hardware resources for the different virtualizedapplications to managing the spectrum or the bandwidth that might be in use forcomputational offloading by different service providers.Blockchain technology can provide multiple advantages to orchestration at theedge: enhanced security and identity management, together with distributed con-sensus algorithms to implement the resource allocation decision processes. In thisarea, EdgeChain was introduced by Zhu et al. as a middleware platform to de-ploy third-party applications across the MEC layer [31]. In [33], the authors intro-duce a blockchain framework that relies on smart contracts for managing networkbandwidth and resource allocation in a distributed and collaborative computationaloffloading framework. In [36], a similar idea is extended towards managing net-work infrastructure and the available computational resources focused at enhanc-ing autonomy of self-driving cars and other autonomous robots forming distributedrobotic systems. In this paper, the blockchain MEC slice was the key slice managingthe deployment of applications across other MEC slices supporting different verti-cals within the automotive sector. Further adoption of blockchain for computationaloffloading will require, however, higher-bandwidth and lower-latency blockchainframeworks enabling real-time sensor data to be streamed for applications such asautonomous mobile robots [30, 106].Resource orchestration processes have underlying optimization algorithms thatcan be implemented either in a more traditional deterministic manner, or relying onmachine learning models. Several authors have proposed the utilization of deep rein-forcement learning for computational offloading in blockchain-powered edge com- lockchain for MEC: Consensus Mechanisms and Scalability 15
UC1:
Blockchain for Edge Resource Orchestration
Pool of Software-Defined Edge Services
Tenant 1 … Tenant N
Application 1 Application 1 … …
Application M Application M N Pool of Edge ResourcesActive Edge Services . . .
End-UsersBlockchain-Managed Resource Allocation and Service Provision
UC2:
Blockchain Marketplace at the Edge
Application 1
Blockchain-Powered Marketplace
Application 2
Application N . . .
UC3:
Blockchain-Enhanced Edge Services (Privacy, Security, Identity Management)
Application 1 Application 2 Application N . . .
Blockchain Blockchain Blockchain
End-Users
Fig. 2: Main use cases for blockchain within edge computing systems. (UC1)Blockchain-powered resource allocation and service provision; (UC2)Blockchain-powered marketplace for interfacing users and services; (UC3)Blockchain-enhanced individual edge services relying on blockchain tech-nology for security, privacy, data management and audits or identity man-agement, among others. puting. In [107], the authors demonstrate an approach that is able to improve long-term performance in a computational offloading scheme, with an adaptive geneticalgorithm to improve the exploration processes while learning. In [43], the authorsdescribe different situations in which blockchain can support resource managementat the edge with deep reinforcement learning: spectrum sharing, vehicle-to-vehicleenergy trading, computational offloading, or device-to-device content caching.
DLT can also provide a platform for building a marketplace between end-users andthird-party edge application through either a transparent, secure and auditable mon-etization framework or as a middleware for sharing data securely between producersand consumers. In the former direction, Xiong et al. deployed a blockchain at theedge to enable resource-constrained devices producing data to sell it to third-partyapplications [32]. The pricing scheme introduced in the paper models the interac-tions within the IoT as market activities and the blockchain represented the frame-work for regulation of such activities. Distributed marketplaces based on blockchainfor MEC services often utilize Ethereum as a base and the InterPlanetary File Sys-tem (IPFS) for data storage. Examples are available in [108] or [109]. A study de-scribing the different challenges and opportunities is available in [110].
In [49], the authors describe how blockchain can play a key enabled role in inter-connected vehicles from the security point of view. In particular, blockchain is ex-ploited for data management, but also for energy management in electric vehicles,with the authors proposing blockchain inspired data coins and energy coins. An edgecomputing security scheme is proposed including these two interaction aspects. Anapproach more related to the nature of blockchain as a cryptocurrency frameworkwas proposed by Liu et al. in [41], where the authors present an offloading frame-work not for data but for the blockchain itself and related mining operations. Ingeneral, blockchain can support edge services by providing enhanced privacy andsecurity [40], decentralized data management [18], or identity management [111].
In this section we describe the benefits and drawbacks of the different consensusprotocols and DLT solutions for each of the three main use cases defined in the pre-vious section and illustrated in Fig. 2, as well as for edge computing in the industrial lockchain for MEC: Consensus Mechanisms and Scalability 17 internet of things. A basic classification of some of the protocols introduced in theprevious section from the point of view of the capabilities of embedded IoT systemsis given in Table 1.Table 1: Comparison of consensus protocols in terms of their applicability withinresource-constrained devices in the IoT.
PoW PoS/DPoS PBFT ConcordiceComputationally-Constrained Devices (cid:55) (cid:51) (cid:55) (cid:51)
Communication-Constrained Devices - (cid:55) (cid:55) (cid:51) Intermittent Connectivity (cid:55) (cid:55) (cid:55) (cid:51)
Independent Subnetworks (cid:55) (cid:55) * (cid:51) ** (cid:51) *** Production-Ready Platform (cid:51) (cid:51) (cid:51) (cid:55) *Recent proposals implementing sharding might be considered subnetworks, however here werefer to the ability of specifically creating a subnetwork from a given set of nodes.*Channels in Hyperledger enable data separation but need to remain connected to the main net.**The tangle in IOTA enables sets of nodes to be disconnected for certain periods of time andrejoin the network later on.
Consensus protocols in the different DLTs are the key performance indicators, andthey are directly related to the minimum capabilities that nodes in the networkmust meet. In PoW-based blockchains, including Bitcoin and Ethereum, resource-constrained devices in the IoT that are potentially battery powered do not havethe ability to participate as full nodes in the network. In Ethereum, nonetheless,the blockchain has adapted to some extent towards embedded IoT devices. For in-stance, the Zerynth Ethereum library provides basic capability to embedded mi-crocontrollers running MicroPython [112]. It enables sensor nodes to create signedtransactions and execute contract calls.Hyperledger Fabric and IOTA, designed with scalability in mind, do not havesuch strong computational requirements. The consensus protocols at the hearth ofHyperledger, however, have high communication complexity and therefore requirenodes to be able to communicate frequently and with low-latency. Hyperledger cantherefore run in embedded IoT edge gateways with wired internet connection but itsextendability to wireless and potentially battery-powered sensor nodes is limited. Inthis area, IOTA has a comparative advantage. In particular, STMicroelectronics hascollaborated with the IOTA foundation in the development of X-CUBE-IOTA [113],a complete middleware that enables IoT sensor nodes based on STM32 microcon-trollers to build IOTA applications and access the IOTA distributed ledger directly.
In terms of communication-constrained devices, low-power wide area networks(LPWANs) have emerged in recent years as a solution for extending the range ofapplications, with LoRa and LoRaWAN being the most prominent radio and net-work technologies [7, 114]. Edge computing is a natural paradigm to be integratedwith LPWAN networks owing to the low-bandwidth available and thus the need topreprocess large amounts of raw data [115, 116, 117]. However, the integration ofblockchain into LPWAN networks is not direct [118]. Current efforts deploy theblockchain either at the LPWAN gateways, which often have wired internet con-nection, or at the back-end servers [119, 120]. More interesting use cases will bepossible when the blockchain nodes can be interconnected via low-bandwidth andhigh-latency LPWAN links, which might be soon possible with IOTA and STM.
From the point of view of edge computing as a system encompassing multiple inde-pendent applications, the simplest use case is such in which blockchains are man-aged by each application independently. This allows for the same orchestration al-gorithms to remain in place, as well as co-existence of blockchain-based and otherapplications running at the edge. Depending on the nature of each of the applica-tions, all of the DLT solutions presented in this chapter might be applied. For generalIoT systems where data is gathered from sensor nodes and transactions between ei-ther the user or the sensors and the application back-end (which may or may notbe deployed entirely at the edge) are relatively simple, then IOTA stands out byproviding free transactions. This can be a key differentiating point in applicationswhere data is routinely gathered and does not have specific value. Because IOTA’sconsensus is built in a way that all nodes need to take the validator role before beingable to commit transactions, nodes do no need an additional incentive to validateand therefore there is no need for a transaction fee as with other blockchain plat-forms. If more complex transactions are required, with either real-time interactionbetween users or a user and sensor data being processed, then smart contracts mightbe needed. Ethereum is by far the most extended and used blockchain platform forsmart contracts, and therefore it would be natural to rely on it. This will be an evenbetter solution when Ethereum 2.0 is available. Nonetheless, relying on Ethereum orsimilar solutions involves an extra transaction cost, due to the need for mining newcryptocurrency to compensate nodes participating in the validation process. Alter-natively, private Ethereum networks can be deployed and infrastructure managed bythe application developers. This is specially important in PoW-based systems, butalso in PoS systems as otherwise nodes would have no incentives on putting theirstakes at risk.When blockchain is utilized to power a marketplace of services at the edge, thecryptocurrency that blockchains build upon might play a more important role withthe introduction of monetization. In this sense, monetization does not necessarilyrefer only to paying for services, but can also encompass the edge resources that lockchain for MEC: Consensus Mechanisms and Scalability 19 services rely on [29]. Similar to the previous case, the choice of DLT frameworkhas a significant dependence on the type of data management and processing thatneeds to be done. For simple applications in which services and end-users are pre-defined and communicate independently, then IOTA can provide a fast and scalableframework, while Hyperledger could be an alternative if there is enough infrastruc-ture set to sustain the blockchain and validate transactions. These applications cancover a wide variety of scenarios: paying a highway toll, exchange of informationfor coordination between autonomous cars, track-and-trace in the logistics sector,or providing digital identity to citizens in a smart city. In all these cases, a commondenominator is that the transfers of value, or information, are small and frequent intime, and therefore there is not enough incentive to utilize other blockchain plat-forms such as Ethereum where transactions involve a fee. Hyperledger, nonetheless,is only a viable option if either public or private infrastructure supports its use with-out an impact on the end-user. For more complex applications, both Hyperledger andEthereum provide extensive support for smart contracts and execution of distributedapplications.The last of the use cases presented in the previous section, and involving themost complex system-level integration of DLT technology at the edge layer is re-source allocation and service provision. In this case, different optimization algo-rithms in which the resource orchestrator relies need to be implemented on top ofthe blockchain for transparent management of resources. The processes involved indynamic resource allocation and service provision are complex and therefore requireblockchains able of running smart contracts. Ethereum provides a suitable platformfrom the functionality point of view, but lacks the ability to scale and the low con-trol over latency would significantly affect the real-time allocation of resources.Moreover, the computational power needed to validate transactions would reducethe availability of edge resources. Until Ethereum 2.0 or a more scalable solution isavailable, Hyperledger has multiple competitive advantages in this area.A different application scenario that has not been directly covered in the previoussection is the industrial IoT. Industrial scenarios often differentiate in that they oper-ate on private networks. Moreover, safety-critical applications require more controlover the network parameters as well as over the data management itself. In thesedirections, Hyperledger Fabric stands out, with design decisions targeting industrialuse cases since its inception. Not only does a permissioned Hyperledger blockchainprovide a secure framework for management of identities and network control, butit is the ability to separate data across channels that can provide wider adoption inprivacy-critical and safety-critical use cases.
We have reviewed the most important consensus protocols in traditional blockchainsand novel distributed ledger technologies, together with the different applicationsand use cases resulting of the integration of blockchain and edge computing. In particular, we have described how the underlying consensus protocols affect the ap-plicability of the different DLT systems for edge computing, with an emphasis onthe current research trends in terms of scalability and performance. We have out-lined the main benefits and drawbacks of Ethereum, Hyperledger and IOTA in fourmain use cases: (i) orchestration of edge resources and services, (ii) implementa-tion of a marketplace of edge services, (iii) enhancing security, privacy or identitymanagement of individual edge services, and (iv) providing a framework for datamanagement in the industrial Internet of Things.
Acknowledgements
This work was supported by the Academy of Finland’s AutoSOS projectwith grant number 328755.
References
1. Li Da Xu, Wu He, and Shancang Li. Internet of things in industries: A survey.
IEEE Trans-actions on industrial informatics , 10(4):2233–2243, 2014.2. Ola Salman, Imad Elhajj, Ali Chehab, and Ayman Kayssi. Iot survey: An sdn and fog com-puting perspective.
Computer Networks , 143:221–246, 2018.3. Celimuge Wu, Zhi Liu, Di Zhang, Tsutomu Yoshinaga, and Yusheng Ji. Spatial intelligencetoward trustworthy vehicular iot.
IEEE Communications Magazine , 56(10):22–27, 2018.4. Jorge Pe˜na Queralta, Tuan Nguyen Gia, Hannu Tenhunen, and Tomi Westerlund. Collabora-tive mapping with ioe-based heterogeneous vehicles for enhanced situational awareness. In , pages 1–6. IEEE, 2019.5. Charalampos Doukas and Ilias Maglogiannis. Bringing iot and cloud computing towards per-vasive healthcare. In , pages 922–926. IEEE, 2012.6. Ammar Awad Mutlag, Mohd Khanapi Abd Ghani, Net al Arunkumar, Mazin Abed Mo-hammed, and Othman Mohd. Enabling technologies for fog computing in healthcare iotsystems.
Future Generation Computer Systems , 90:62–78, 2019.7. Jorge Pe˜na Queralta, Tuan Nguyen Gia, Hannu Tenhunen, and Tomi Westerlund. Edge-AI in LoRa based healthcare monitoring: A case study on fall detection system with LSTMRecurrent Neural Networks. In , 2019.8. Sam Edwards and Ioannis Profetis. Hajime: Analysis of a decentralized internet worm foriot devices.
Rapidity Networks , 16, 2016.9. Yun Chao Hu, Milan Patel, Dario Sabella, Nurit Sprecher, and Valerie Young. Mobile edgecomputinga key technology towards 5g.
ETSI white paper , 11(11):1–16, 2015.10. Weisong Shi, Jie Cao, Quan Zhang, Youhuizi Li, and Lanyu Xu. Edge computing: Visionand challenges.
IEEE internet of things journal , 3(5):637–646, 2016.11. L. Qingqing, F. Yuhong, J. Pe˜na Queralta, T. N. Gia, Z. Zou, H. Tenhunen, T. Westerlund.Edge Computing for Mobile Robots: Multi-Robot Feature-Based Lidar Odometry with FP-GAs. In . IEEE, 2019.12. Melanie Swan.
Blockchain: Blueprint for a new economy . ” O’Reilly Media, Inc.”, 2015.13. Sarah Underwood. Blockchain beyond bitcoin, 2016.14. Yang Lu. Industry 4.0: A survey on technologies, applications and open research issues.
Journal of Industrial Information Integration , 6:1–10, 2017.15. Yongfeng Qian, Yingying Jiang, Jing Chen, Yu Zhang, Jeungeun Song, Ming Zhou, andMatevˇz Pustiˇsek. Towards decentralized iot security enhancement: A blockchain approach.
Computers & Electrical Engineering , 72:266–273, 2018.lockchain for MEC: Consensus Mechanisms and Scalability 2116. Juah C Song, Mevlut A Demir, John J Prevost, and Paul Rad. Blockchain design for trusteddecentralized iot networks. In , pages 169–174. IEEE, 2018.17. Mohamed Tahar Hammi, Badis Hammi, Patrick Bellot, and Ahmed Serhrouchni. Bubbles oftrust: A decentralized blockchain-based authentication system for iot.
Computers & Security ,78:126–142, 2018.18. Gbadebo Ayoade, Vishal Karande, Latifur Khan, and Kevin Hamlen. Decentralized iot datamanagement using blockchain and trusted execution environment. In , pages 15–22. IEEE, 2018.19. Jollen Chen. Devify: Decentralized internet of things software framework for a peer-to-peerand interoperable iot device.
ACM SIGBED Review , 15(2):31–36, 2018.20. Penn H Su, Chi-Sheng Shih, Jane Yung-Jen Hsu, Kwei-Jay Lin, and Yu-Chung Wang. De-centralized fault tolerance mechanism for intelligent iot/m2m middleware. In , pages 45–50. IEEE, 2014.21. Zibin Zheng, Shaoan Xie, Hongning Dai, Xiangping Chen, and Huaimin Wang. An overviewof blockchain technology: Architecture, consensus, and future trends. In , pages 557–564. IEEE, 2017.22. Marko Vukoli´c. The quest for scalable blockchain fabric: Proof-of-work vs. bft replication.In
International workshop on open problems in network security . Springer, 2015.23. Ghassan Karame. On the security and scalability of bitcoin’s blockchain. In
Proceedings ofthe 2016 ACM SIGSAC conference on computer and communications security , pages 1861–1862, 2016.24. Mattias Scherer. Performance and scalability of blockchain networks and smart contracts,2017.25. Loi Luu, Viswesh Narayanan, Chaodong Zheng, Kunal Baweja, Seth Gilbert, and PrateekSaxena. A secure sharding protocol for open blockchains. In
Proceedings of the 2016 ACMSIGSAC Conference on Computer and Communications Security , pages 17–30, 2016.26. Eleftherios Kokoris-Kogias, Philipp Jovanovic, Linus Gasser, Nicolas Gailly, Ewa Syta, andBryan Ford. Omniledger: A secure, scale-out, decentralized ledger via sharding. In , pages 583–598. IEEE, 2018.27. Sami Kekki, Walter Featherstone, Yonggang Fang, Pekka Kuure, Alice Li, Anurag Ranjan,Debashish Purkayastha, Feng Jiangping, Danny Frydman, Gianluca Verin, et al. Mec in 5gnetworks.
ETSI white paper , 28:1–28, 2018.28. Sonia Shahzadi, Muddesar Iqbal, Tasos Dagiuklas, and Zia Ul Qayyum. Multi-access edgecomputing: open issues, challenges and future perspectives.
Journal of Cloud Computing ,6(1):30, 2017.29. Tarik Taleb, Konstantinos Samdanis, Badr Mada, Hannu Flinck, Sunny Dutta, and DarioSabella. On multi-access edge computing: A survey of the emerging 5g network edge cloudarchitecture and orchestration.
IEEE Communications Surveys & Tutorials , 19(3), 2017.30. L. Qingqing, J. Pe˜na Queralta, T. N. Gia, Z. Zou, H. Tenhunen, T. Westerlund. Visual Odom-etry Offloading in Internet of Vehicles with Compression at the Edge of the Network. In , 2019.31. He Zhu, Changcheng Huang, and Jiayu Zhou. Edgechain: Blockchain-based multi-vendormobile edge application placement. In , pages 222–226. IEEE, 2018.32. Zehui Xiong, Yang Zhang, Dusit Niyato, Ping Wang, and Zhu Han. When mobile blockchainmeets edge computing.
IEEE Communications Magazine , 56(8):33–39, 2018.33. Jorge Pe˜na Queralta and Tomi Westerlund. Blockchain-powered collaboration in heteroge-neous swarms of robots.
Frontiers in Robotics and AI (to appear) , 2020. Presented at theSymposium on Blockchain for Robotic and AI Systems, MIT Media Lab.34. MD Abdur Rahman, M Shamim Hossain, George Loukas, Elham Hassanain, Syed SadiqurRahman, Mohammed F Alhamid, and Mohsen Guizani. Blockchain-based mobile edge com-puting framework for secure therapy applications.
IEEE Access , 6:72469–72478, 2018.35. 3GPP. Study on architecture for next-generation system rel. 14.
Techical Report , 2016.2 Jorge Pe˜na Queralta and Tomi Westerlund36. Jorge Pe˜na Queralta, Li Qingqing, Zhuo Zou, and Tomi Westerlund. Enhancing autonomywith blockchain and multi-acess edge computing in distributed robotic systems. In
The FifthInternational Conference on Fog and Mobile Edge Computing (FMEC) . IEEE, 2020.37. N. Alliance. Description of network slicing concept.
NGMN 5G P , 1:1, 2016.38. Fabio Giust, Vincenzo Sciancalepore, Dario Sabella, Miltiades C Filippou, Simone Man-giante, Walter Featherstone, and Daniele Munaretto. Multi-access edge computing: Thedriver behind the wheel of 5g-connected cars.
IEEE Communications Standards Magazine ,2(3):66–73, 2018.39. Roberto Casado-Vara, Fernando de la Prieta, Javier Prieto, and Juan M Corchado. Blockchainframework for iot data quality via edge computing. In
Proceedings of the 1st Workshop onBlockchain-enabled Networked Sensor Systems , pages 19–24, 2018.40. A. Nawaz, J. Pe˜na Queralta, T. N. Gia, H. Kan, T. Westerlund. Edge AI and Blockchain forPrivacy-Critical and Data-Sensitive Applications. In
The 12th International Conference onMobile Computing and Ubiquitous Networking (ICMU) , 2019.41. Mengting Liu, F Richard Yu, Yinglei Teng, Victor CM Leung, and Mei Song. Joint compu-tation offloading and content caching for wireless blockchain networks. In
IEEE INFOCOM2018-IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS) ,pages 517–522. IEEE, 2018.42. Jorge Pe˜na Queralta, Li Qingqing, Tuan Nguyen Gia, Hong-Linh Truong, and Tomi Wester-lund. End-to-end design for self-reconfigurable heterogeneous robotic swarms. In
The 16thInternational Conference on Distributed Computing in Sensor Systems . IEEE, 2020.43. Yueyue Dai, Du Xu, Sabita Maharjan, Zhuang Chen, Qian He, and Yan Zhang. Blockchainand deep reinforcement learning empowered intelligent 5g beyond.
IEEE Network , 33, 2019.44. Nguyen Cong Luong, Zehui Xiong, Ping Wang, and Dusit Niyato. Optimal auction for edgecomputing resource management in mobile blockchain networks: A deep learning approach.In , pages 1–6. IEEE, 2018.45. Mayra Samaniego and Ralph Deters. Hosting virtual iot resources on edge-hosts withblockchain. In , pages 116–119. IEEE, 2016.46. Mayra Samaniego and Ralph Deters. Using blockchain to push software-defined iot com-ponents onto edge hosts. In
Proceedings of the International Conference on Big Data andAdvanced Wireless Technologies , pages 1–9, 2016.47. Mayra Samaniego and Ralph Deters. Virtual resources & blockchain for configuration man-agement in iot.
Journal of Ubiquitous Systems & Pervasive Networks , 9(2):1–13, 2017.48. The European Union Agency for Cybersecurity. Threat assessment for the fifth generationof mobile telecommunications networks (5g).
ENISA , 2019.49. Hong Liu, Yan Zhang, and Tao Yang. Blockchain-enabled security in electric vehicles cloudand edge computing.
IEEE Network , 32(3):78–83, 2018.50. Jiawen Kang, Rong Yu, Xumin Huang, Maoqiang Wu, Sabita Maharjan, Shengli Xie, andYan Zhang. Blockchain for secure and efficient data sharing in vehicular edge computingand networks.
IEEE Internet of Things Journal , 6(3):4660–4670, 2018.51. T. N. Gia, A. Nawaz, J. Pe˜na Queralta, T. Westerlund. Artificial Intelligence at the Edge inthe Blockchain of Things. In , 2019.52. Eduardo Castell´o Ferrer, Ognjen Rudovic, Thomas Hardjono, and Alex Pentland.Robochain: A secure data-sharing framework for human-robot interaction. arXiv preprintarXiv:1802.04480 , 2018.53. Ruizhe Yang, F Richard Yu, Pengbo Si, Zhaoxin Yang, and Yanhua Zhang. Integratedblockchain and edge computing systems: A survey, some research issues and challenges.
IEEE Communications Surveys & Tutorials , 21(2):1508–1532, 2019.54. Pietro Danzi, Anders E Kalør, ˇCedomir Stefanovi´c, and Petar Popovski. Delay and com-munication tradeoffs for blockchain systems with lightweight iot clients.
IEEE Internet ofThings Journal , 6(2):2354–2365, 2019.lockchain for MEC: Consensus Mechanisms and Scalability 2355. Seyednima Khezr, Md Moniruzzaman, Abdulsalam Yassine, and Rachid Benlamri.Blockchain technology in healthcare: A comprehensive review and directions for future re-search.
Applied Sciences , 9(9):1736, 2019.56. Dinh C Nguyen, Pubudu N Pathirana, Ming Ding, and Aruna Seneviratne. Blockchain for5g and beyond networks: A state of the art survey.
Journal of Network and Computer Appli-cations , page 102693, 2020.57. Weichao Gao, William G Hatcher, and Wei Yu. A survey of blockchain: techniques, applica-tions, and challenges. In , pages 1–11. IEEE, 2018.58. Archana Prashanth Joshi, Meng Han, and Yan Wang. A survey on security and privacy issuesof blockchain technology.
Mathematical Foundations of Computing , 1(2):121–147, 2018.59. Wazir Zada Khan, Ejaz Ahmed, Saqib Hakak, Ibrar Yaqoob, and Arif Ahmed. Edge com-puting: A survey.
Future Generation Computer Systems , 97:219–235, 2019.60. Jose Moura and David Hutchison. Fog computing systems: State of the art, research issuesand future trends. arXiv preprint arXiv:1908.05077 [v2] , pages 1–32, 2020.61. Xiaoqi Li, Peng Jiang, Ting Chen, Xiapu Luo, and Qiaoyan Wen. A survey on the securityof blockchain systems.
Future Generation Computer Systems , 2017.62. Satoshi Nakamoto et al.
Bitcoin: A peer-to-peer electronic cash system . 2008.63. Cynthia Dwork and Moni Naor. Pricing via processing or combatting junk mail. In
AnnualInternational Cryptology Conference , pages 139–147. Springer, 1992.64. Giang-Truong Nguyen and Kyungbaek Kim. A survey about consensus algorithms used inblockchain.
Journal of Information processing systems , 14(1), 2018.65. Sunny King. Primecoin: Cryptocurrency with prime number proof-of-work. 1:6, 2013.66. Rasmus Pagh and Flemming Friche Rodler. Cuckoo hashing.
Journal of Algorithms ,51(2):122–144, 2004.67. Simon Barber, Xavier Boyen, Elaine Shi, and Ersin Uzun. Bitter to betterhow to make bitcoina better currency. In
Financial Cryptography and Data Security . Springer, 2012.68. Serguei Popov. A probabilistic analysis of the nxt forging algorithm.
Ledger , 1:69–83, 2016.69. Nxt Wiki.
Whitepaper: Nxt . Nxtwiki. org [online] https://nxtwiki. org, 2018.70. Iddo Bentov, Ariel Gabizon, and Alex Mizrahi. Cryptocurrencies without proof of work. In
International Conference on Financial Cryptography and Data Security . Springer, 2016.71. Karl J O’Dwyer and David Malone.
Bitcoin mining and its energy footprint . IET, 2014.72. Alex De Vries. Bitcoin’s growing energy problem.
Joule , 2(5):801–805, 2018.73. Vitalik Buterin, Daniel Reijsbergen, Stefanos Leonardos, and Georgios Piliouras. Incentivesin ethereum’s hybrid casper protocol. arXiv preprint arXiv:1903.04205 , 2019.74. Sunny King and Scott Nadal. Ppcoin: Peer-to-peer crypto-currency with proof-of-stake. self-published paper, August , 19, 2012.75. Miguel Castro, Barbara Liskov, et al. Practical byzantine fault tolerance. In
OSDI , volume 99,pages 173–186, 1999.76. RM Keichafer, Chris J. Walter, Alan M. Finn, and Philip M. Thambidurai. The maft archi-tecture for distributed fault tolerance.
IEEE Transactions on Computers , 37(4), 1988.77. Joao Sousa, Alysson Bessani, and Marko Vukolic. A byzantine fault-tolerant ordering servicefor the hyperledger fabric blockchain platform. In , pages 51–58. IEEE, 2018.78. C. Cachin. Architecture of the hyperledger blockchain fabric. In
Workshop on distributedcryptocurrencies and consensus ledgers , volume 310, page 4, 2016.79. Chinmay Saraf and Siddharth Sabadra. Blockchain platforms: A compendium. In , pages1–6. IEEE, 2018.80. Elli Androulaki, Artem Barger, Vita Bortnikov, Christian Cachin, Konstantinos Christidis,Angelo De Caro, David Enyeart, Christopher Ferris, Gennady Laventman, Yacov Manevich,et al. Hyperledger fabric: a distributed operating system for permissioned blockchains. In
Proceedings of the Thirteenth EuroSys Conference , pages 1–15, 2018.81. Diego Ongaro and John Ousterhout. In search of an understandable consensus algorithm.In { USENIX } Annual Technical Conference ( { USENIX }{ ATC } , pages 305–319,2014.4 Jorge Pe˜na Queralta and Tomi Westerlund82. Pierre-Louis Aublin, Sonia Ben Mokhtar, and Vivien Qu´ema. Rbft: Redundant byzantinefault tolerance. In , pages 297–306. IEEE, 2013.83. S Popov. The tangle, iota whitepaper. Technical report, IOTA, Tech. Rep.[Online]. Available:https://iota. org/IOTA Whitepaper. pdf, 2018.84. M Divya and Nagaveni B Biradar. Iota-next generation block chain. International JournalOf Engineering And Computer Science , 7(04):23823–23826, 2018.85. Serguei Popov, Hans Moog, Darcy Camargo, Angelo Capossele, Vassil Dimitrov, Alon Gal,Andrew Greve, Bartosz Kusmierz, Sebastian Mueller, Andreas Penzkofer, et al. The coordi-cide, 2020.86. Serguei Popov and William J Buchanan. Fpc-bi: Fast probabilistic consensus within byzan-tine infrastructures. arXiv preprint arXiv:1905.10895 , 2019.87. Daniel Ramos and Gabriel Zanko. Review of iota foundation as a moving force for massiveblockchain adoption in different industry sectors.88. KENRIC NELSON and ANDR ´E VILELA. Majority vote dynamics for iota transactionconsensus. 2020.89. Colin LeMahieu. Nano: A feeless distributed cryptocurrency network.
Nano [Online re-source]. URL: https://nano. org/en/whitepaper (date of access: 24.03. 2018) , 2018.90. Skycoin.com. Skycoin whitepaper v1.2. Technical report, [Online]. Available:https://downloads.skycoin.com/whitepapers/Skycoin-Whitepaper-v1.2.pdf, 2020.91. Gavin Wood et al. Ethereum: A secure decentralised generalised transaction ledger.
Ethereum project yellow paper , 151(2014):1–32, 2014.92. Everett Hildenbrandt, Manasvi Saxena, Nishant Rodrigues, Xiaoran Zhu, Philip Daian,Dwight Guth, Brandon Moore, Daejun Park, Yi Zhang, Andrei Stefanescu, et al. Kevm:A complete formal semantics of the ethereum virtual machine. In , pages 204–217. IEEE, 2018.93. Ethereum Revision 7709ece9.
Solidity Documentation . Solidity Read The Docs [online]https://solidity.readthedocs.io/en/v0.5.12/., 2016-2019.94. Chris Dannen.
Introducing Ethereum and Solidity . Springer, 2017.95. Dejan Vujiˇci´c, Dijana Jagodi´c, and Siniˇsa Randi´c. Blockchain technology, bitcoin, andethereum: A brief overview. In , pages 1–6. IEEE, 2018.96. Deadalnix’s den.
Using Merklix tree to shard block validation . [online]https://deadalnix.me/2016/11/06/, 2016.97. Deadalnix’s den.
Introducing Merklix tree as an unordered Merkle tree on steroid
Future Generation Computer Systems , 2017.99. C. Ferris. does hyperledger fabric perform at scale?
Blockchain Pulse: IBM Blockchain Blog ,2, 2019.100. Vitalik Buterin et al. A next-generation smart contract and decentralized application plat-form. white paper , 3:37, 2014.101. Serenity Ethereum Foundation et al.
Ethereum 2.0 Specifications . [online]https://github.com/ethereum/eth2.0-specs, 2018.102. Vitalik Buterin and Virgil Griffith. Casper the friendly finality gadget. arXiv preprintarXiv:1710.09437 , 2017.103. Seyoung Huh, Sangrae Cho, and Soohyung Kim. Managing iot devices using blockchainplatform. In , pages 464–467. IEEE, 2017.104. Matevˇz Pustiˇsek and Andrej Kos. Approaches to front-end iot application development forthe ethereum blockchain.
Procedia Computer Science , 129:410–419, 2018.105. Martin Valenta and Philipp Sandner. Comparison of ethereum, hyperledger fabric and corda. no. June , pages 1–8, 2017.106. Li Qingqing, Jorge Pe˜na Queralta, Tuan Nguyen Gia, and Tomi Westerlund. OffloadingMonocular Visual Odometry with Edge Computing: Optimizing Image Compression Ratiosin Multi-Robot Systems. In
The 5th International Conference on Systems, Control and Com-munications (ICSCC) , 2019.lockchain for MEC: Consensus Mechanisms and Scalability 25107. Xiaoyu Qiu, Luobin Liu, Wuhui Chen, Zicong Hong, and Zibin Zheng. Online deep rein-forcement learning for computation offloading in blockchain-empowered mobile edge com-puting.
IEEE Transactions on Vehicular Technology , 68(8):8050–8062, 2019.108. Kazim Rifat ¨Ozyilmaz, Mehmet Do˘gan, and Arda Yurdakul. Idmob: Iot data marketplace onblockchain. In
Crypto Valley Conference on Blockchain Technology (CVCBT) . IEEE, 2018.109. Vishnu Prasad Ranganthan, Ram Dantu, Aditya Paul, Paula Mears, and Kirill Morozov. Adecentralized marketplace application on the ethereum blockchain. In . IEEE, 2018.110. Blesson Varghese, Massimo Villari, Omer Rana, Philip James, Tejal Shah, Maria Fazio, andRajiv Ranjan. Realizing edge marketplaces: challenges and opportunities.
IEEE Cloud Com-puting , 5(6):9–20, 2018.111. Yongjun Ren, Fujian Zhu, Jian Qi, Jin Wang, and Arun Kumar Sangaiah. Identity manage-ment and access control based on blockchain under edge computing for the industrial internetof things.
Applied Sciences
Procedia Computer Science , 2019.115. V. K. Sarker, J. Pe˜na Queralta, T. N. Gia, H. Tenhunen, T. Westerlund. A survey on lora foriot: Integrating edge computing. In
SLICE- FMEC , 2019.116. T. N. Gia, L. Qingqing, J. Pe˜na Queralta, H. Tenhunen, T. Westerlund. Edge AI in SmartFarming IoT: CNNs at the Edge and Fog Computing with LoRa. In
IEEE AFRICON , 2019.117. T. N. Gia, J. Pe˜na Queralta, T. Westerlund. Exploiting LoRa, Edge and Fog Computing forTraffic Monitoring in Smart Cities. In
Book Chapter: LPWAN Technologies for IoT and M2MApplications . Elsevier, 2020.118. Kazım Rıfat ¨Ozyılmaz and Arda Yurdakul. Work-in-progress: integrating low-power iotdevices to a blockchain-based infrastructure. In , pages 1–2. IEEE, 2017.119. Jun Lin, Zhiqi Shen, Chunyan Miao, and Siyuan Liu. Using blockchain to build trustedlorawan sharing server.
International Journal of Crowd Science , 2017.120. Arnaud Durand, Pascal Gremaud, and Jacques Pasquier. Resilient, crowd-sourced lpwaninfrastructure using blockchain. In