INTRODUCTION

A security framework is a set of functions, technologies, and protocols that provides a secure foundation for the implementation of applications. Using a security framework improves the system’s optimization and ensures the accuracy and reliability of the stored data. A framework for an IoT network does not depend on any types of IoT devices in the network and can easily integrate new modules without changing the IoT network architecture.

Currently, the security frameworks for IoT networks can be divided into two main groups: centralized frameworks and decentralized frameworks. In centralized frameworks, a central node is responsible for storing data and providing services to the network, and other nodes can send requests to this node. Some typical solutions of this type are introduced in1, 2, 3. However, these solutions have three limitations as follows4, 5: (1) : all data stored in the central system can be altered or removed by any people controlling the system; (2) : all nodes cannot be able to access the services if the central system is stopped operation due to overload, denial-of-service or distributed denial-of-service attacks, or system errors; (3) when the number of IoT devices and resources increases dramatically, the tasks of administration, configuration, and scalability become more complex.

For decentralized frameworks, most solutions use BC technology as the main component in the systems due to its advantages such as , , , and 6, 7, 8. Moreover, an IoT network is usually owned by an exclusive organization. Hence, the private BC model is a more suitable selection for improving security and reducing network delay compared to the public BC model. However, the current consensus protocols used in these frameworks are not yet optimized resources for miners. Particularly, there are two assumptions about miners in a private BC as follows: (1) all miners in a private BC network are trusted nodes. However, the speed verification for transactions remains unchanged when adding some miners to the network, and (2) The assumption that a BC network exists some untrusted miners but less than one-third of the total number of miners, in case a malicious miner is selected at a mining round, it absolutely can put some invalid transactions in a new block and then broadcasts that block to the network. Of course, this invalid block is dropped by other miners; however, the valid transactions in this block have to be reverified in a future mining round. Therefore, we proposed a BC-based security framework for IoT networks, which aims to guarantee the reliability of data in the BC ledger and improve miners’ performance.

The objectives of the proposed framework are as follows:

The proposed framework includes two phases: the verification phase and the block making phase. We also consider two assumptions when all BC miners are trusted nodes and some miners are untrusted nodes. Our contribution in this paper can be summarized as follows: (1) We propose a novel security framework based on BC for IoT networks; (2) We compare the proposed framework with other decentralized frameworks with the two assumptions discussed above; (3) We evaluate our proposed framework in terms of the mining speed and the number of transactions verified within certain times.

Paper outline: Section II presents the related work of BC-based security frameworks for IoT. The preliminaries are given in Section III. The proposed security framework is described in Section IV. The evaluation is given in Section V. Finally, the paper is concluded in Section VI.

RELATED WORK

Currently, BC-based security frameworks for the IoT provide three main features: access control, authentication and communication, and data security.

Concerning access control, a framework called FairAccess was introduced by Ouaddah et al.9 as a privacy-preserving access control system in the IoT, where smart contracts are used to enforce access control policies and to distribute access tokens to the BC. A requester can use a token issued by a resource owner to access a licensed resource. In addition, each resource owner can also perform a GrantAccess transaction to revoke or update permissions granted to a requester. However, the authors in10 specified the limitations of FairAccess in the high time cost related to distributing authorization to requesters and the lack of integration of the policies with a relationship network. To overcome these limitations, researchers also proposed a solution called ControlChain, which uses four different BCs, named Context BC, Relationships BC, Rules BC, and Accountability BC, to establish relationships between objects and then assign attributes to these relationships.

Concerning authentication and secure communication, Panda et al. 11 proposed an authentication framework for IoT systems. In this proposal, the gateway nodes joined in the BC network are an interface between IoT devices and the Ethereum BC network. Moreover, those nodes are also responsible for controlling connections and translating messages between IoT devices and users. A user can send his/her permission ticket emitted by a smart contract to the gateway to access a certain IoT device. In another work 12, the authors built a decentralized and scalable security framework to provide a secure communication mechanism for users and applications in an IoT environment. The tree-based hash method is responsible for verifying and authenticating messages exchanged between devices, while the BC ledger is used to store and distribute data securely. Similarly, Muhidul Islam Khan et al. 13 introduced a framework for secured communication in a decentralized IoT network. In this framework, a sending node has to sign a contract created by a selected miner node to transfer data to another node. They used a hybrid consensus algorithm consisting of binary and average consensus mechanisms for miner selection in the BC network.

Concerning IoT data security, the BC-based framework addressed in 14 allows both data owners and customers to verify data integrity for IoT data stored in the semitrusted cloud. In this approach, each data block is identified by a hash value of its content before uploading it to a cloud service provider. The encrypted hash value and the corresponding data block ID are written in a smart contract. In15, the authors described a BC-based framework for IoT data trade, in which the processes of device management, data exchange between producers and consumers, and service quality assessment are performed by smart contracts.

The decentralized frameworks discussed above have some important contributions to the IoT domain. However, most of these frameworks focus on providing security services for IoT without mentioning the optimization issue in consensus. Particularly, the authors in9, 15 do not specify a particular consensus protocol in their BC network. The frameworks11, 14 use the proof-of-work (PoW) consensus protocol of the Ethereum BC platform to guarantee the reliability and accuracy of data stored in the BC ledger. Therefore, the mining process requires considerable electricity consumption and computational power for miners 16, 17. In 13, the hybrid consensus will operate like Proof-of-Stake (PoS) when the top miners maintain stable operation, and a selected miner has to take an extra step for sending contracts to the sending nodes for signing. In 7, the authors compare consensus algorithms and indicate that the PBFT, Tendermint, and DPOS protocols are suitable for consortium BCs or private BCs. He Yi et al.18 also propose using Tendermint for a private BC. Deepak Puthal et al. 19 proposed a consensus algorithm named Proof-of-Authentication (PoAh) for a lightweight BC for IoT.

PRELIMINARIES

Blockchain

A BC can be viewed as a linked list of blocks of transactions, and each block refers to its parent block via a hash pointer. The first block of the chain is called the genesis block, which has no parent block, so the value of the hash pointer is initialized by the constructor of the BC network. The structure of each block includes two main components: the block header contains management information of the block and chain, such as Version, Previous block hash, Timestamp, Merkle root, and Nonce, and the block body holds a list of transactions6, as shown in Figure 1.

The number of transactions in each block depends on the size of each transaction and block. However, a BC network will not work efficiently when blocks contain large data because the data synchronization time on the ledger will be very slow due to the network latency, which consumes considerable computational power of miners 20, 21. For Bitcoin electronic currency, each transaction has an average size of 250 bytes, and the maximum block size is set at 1 megabyte. The average block size of the last 100 blocks of Ethereum, measured by the authors in22, is 2.9 kilobytes. There are two types of nodes on a BC network:

- : The nodes only conduct transactions.

- : The nodes are responsible for verifying transactions, creating new blocks, and holding the ledger. A miner node can also perform transactions as normal nodes.

Each node of a BC network is initialized to a public/private key pair where the private key is used to create the signature on transactions while the public key is used to verify the digital signature. In general, there are three types of BC networks: public BC, private BC, and consortium BC23.

PROPOSED FRAMEWORK

Let be the set of miner nodes in a BC network; each miner is responsible for verifying transactions and creating new blocks. Normally, these nodes have a high performance, such as servers or workstations, and own a private key and a corresponding public key in which the public key is considered a node identification or a wallet address on the BC network, while the private key is used to sign on transactions/blocks. All miner nodes must participate in the process of verifying transactions, and one of them is randomly selected for proposing a new block for each mining round. The waiting list is the set of unconfirmed transactions received from IoT devices, which is considered a public transaction pool for miners. Furthermore, the verification list is the set of verified transactions, and the values of and fluctuate from time to time. Let be the maximum number of transactions in each block.

Each IoT device is a node on the BC network and has a key pair similar to a miner node, which can perform transactions. The architecture of the proposed framework is shown in Figure 2. The rules are applied for the VL and WL, including (1) Unverified transactions are saved to WL; (2) Only miners can verify transactions in the WL; (3) If a transaction is confirmed successfully, it will be moved from the WL to the VL otherwise dropped; (4) After adding a new block to the chain successfully, the corresponding transactions in the VL are removed by a system application.

We consider two assumptions about miners on a private BC network as follows:

The processes of creating and synchronizing data on the ledger of the proposed framework consist of two phases as follows:

- Phase 1: The verification phase

When an IoT device makes a transaction which is broadcasted to the BC network and is stored in the tail of the WL. The process of verifying transactions in the WL depends on the two assumptions.

Since all miners are completely trusted, only needs to be verified by one miner in the network, for instance, . If is a valid transaction, it will be added to the tail of VL; otherwise, it will be dropped.

The needs to be verified by at least two-thirds of the miners to be added to the VL.

- Phase 2: Block-making phase

The basic idea is that a selected miner puts the head transactions of the VL into a new block and then signs and transmits that block to other miners for verification. The block is added to the chain if it satisfies the requirements of the particular assumptions. The detailed steps of this phase are as follows:

The goal of this step is to select a miner for creating and broadcasting a new block to the network.

Since the miners are all honest, the administrator can select a fixed miner node and can also change by another node as needed.

A miner is randomly selected for each mining round: , where .

The takes transactions in the VL into a new block, where the value depends on the size of each transaction and the maximum allowed size of each block. Then, generates a digital signature on the new block.

The broadcasts the new block and signature to the other miners. In addition, also adds that block to its local chain.

After receiving a new block, other miners verify the validation of the block proposed by the .

Other miners verify the signature of ; if the signature is valid, the block is added to their chain.

Including the two following substeps

Other miners verify the signature of the . If the verification is successful, the miners go to the next substep; otherwise, the miners drop that block and go to .

Other miners check whether the transactions in the new block are the same as the transactions in the VL. If it is true, the block is added to their chain; otherwise, the miners drop that block and go to .

EVALUATION

In this section, we evaluate the proposed framework in the two assumptions in terms of the mining time of a block containing transactions and the number of verified transactions in a certain time. The current decentralized security frameworks for IoT are mainly focused on providing particular features such as access control, authentication, secure communication, and data security. However, the mechanisms of verifying transactions, creating blocks, and synchronizing data in the ledger depend entirely on consensus protocols. Therefore, we compare the proposed framework to other consensus protocols that can be used in the frameworks mentioned in the related work section.

We assume that all miners have the same computational performance. Let t be the verification time per transaction of a miner, t be the time of generating a digital signature/vote/certificate of a miner for a block, and t be the time of verifying a signature/vote/certificate of a miner. The time of transferring a block/vote/certificate/transaction to a destination (such as the VL or a miner) is t, and t be the time of randomly selecting a miner at each mining round.

The proposed framework in assumption 1

In phase 1, each transaction is verified by only one of the n miners; therefore, the verification time of l transactions is , and the time for transferring l transactions verified by the n miners to the VL is . In phase 2, the time of the first operation is ignored because a miner is fixed for creating blocks; hence, the total time of this phase includes the time at steps 2, 3, and 4, that is, . The total time of generating a new block of our framework, denoted by , is considered the average mining time of the system, which is as follows:

Consider the time of generating a new block of other frameworks using consensus protocols such as PoW, PoS, PoA, PoAh, PBFT, and Tendermint. The common characteristic of these consensus protocols is that a selected miner is responsible for proposing a new block, and other nodes must verify this proposed block. With assumption 1, all miners are trusted nodes and have to publish their public key to the BC network. A selected miner has to sign on a new block before distributing that block to other miners. The general algorithm of these consensus protocols, denoted by A1, is illustrated in Table 1.

Algorithm A1
	Input: l transactions in the pool, a selected miner mi, and other miners. Output: a new block in the BC ledger.
Step 1	(1) mi verifies l transactions and puts them into a new block. (2) mi generates a signature on this block. (3) mi broadcasts that block to other miners as well as adds that block to its local chain.
Step 2	Other miners verify the signature on the received block, if the signature is valid, the block will be added to the chain.

To verify l transactions in , spends . The times of creating a signature and broadcasting the new block at and are t and t, respectively. The time for verifying a signature at Step 2 is t. The total average mining time of the A1, denoted by T’, is calculated as follows:

Let t = 1, t = 1, t = 1, t = 1, and l = 40. Figure 2 shows the average mining times of A1 and the proposed framework. From this figure, it is shown that the mining time of the proposed framework is much lower than.A1. Specifically, the average mining time of A1 T'= 43 when the number of miners n = 10 remains unchanged when increasing the number of miners from 10 to 50, whereas the average mining time of the proposed framework T = 11 when n = 10, and for n = 50, then T = 4.6. This means that the more miners that join the network, the lower the average mining time.

Another experimental analysis is performed in terms of adjusting the parameters n and l in formula (1) of the proposed framework, as shown in Figure 4. As shown in the figure, it is clear that with the same number of miner nodes, the average mining time increases proportionally with the number of transactions in a block. In addition, when the number of miners increases, the average mining time is reduced. Particularly, when the same value n = 10, with the number of transactions in each block l = 10, then the average mining time T = 5; for l = 20, then T = 7; for l = 30, then T = 9; and with l = 40, then T = 11. When increasing the number of miners n = 50, for l = 10 then T = 3.4, for l = 20 then T = 3.8, for l = 30 then T = 4.2, and with l = 40 T = 4.6.

Figure 5 shows that if both solutions are configured with the same average mining time and the number of miners is 20 (n = 20), the total verified transactions of the proposed framework are also significantly higher than the A1. When the average mining time is 5, the numbers of processed transactions of A1 and the proposed framework are 5 and 20, respectively. However, when T is set to 25, the number of processed transactions of A1 increases slightly to 22, while the proposed framework handles up to 220 transactions.

The number of verified transactions of the proposed framework increases proportionally with the verification time and the number of miners on the network, as shown in Figure 6. In the same mining time, the system with more miners has more verified transactions than other systems. Specifically, considering the average mining time T = 5, when the number of miners in the network n = 5, the total number of verified transactions from these miners l = 5; when n = 10, then l = 10, and n = 15 corresponds to l = 15. Consider T = 40; for n = 5, then l ≈ 93 transactions; for n = 10, then l = 185; and for n = 15, l ≈ 278.

Our evaluation results show that using the proposed security framework in this assumption is well suited for private BC networks, which will be highly effective in the process of transactions verification and data consensus on the BC ledger. However, all miners in the private BC network must be completely trusted, which means that they are hardly compromised by any attackers and do not commit any fraud in the BC network.

The proposed framework in assumption 2

Each transaction must be verified by at least two-thirds of all miners, which means that a valid transaction will be issued a certificate by a trusted miner. If a transaction has reached two-thirds of certificates, it will be immediately moved to the VL by a monitor application without further verification; otherwise, it will be removed from the WL. In the case of a network with many untrusted miners but less than one-third of the total number of miners, each transaction has to be verified by all miners. This work is similar to other consensus protocols when considering that all miners are the same as a miner, as shown in Figure 7.

In this assumption, we compare the proposed framework to the PBFT and Tendermint protocols. Tendermint is based on PBFT consensus, and both of these consensuses operate based on the assumption that less than one-third of the miners are untrusted nodes. We describe the general three steps of these two protocols, called A2, as shown in Table 2.

Algorithm A2
	Input: l transactions in the pool and miners Output: a new block in the BC ledger
Step 1	Randomly select one miner at each mining round (e.g. mi) mi verifies puts l valid transactions into a new block. mi creates a vote on this block. mi broadcasts this block along with a vote on the proposed block to other miners on the BC network.
Step 2	Each miner has to perform the following works after receiving a new block and vote from mi: Verifying transactions in this block. If all transactions are valid, the miner generates its vote for that block. Broadcasting the vote to other miners. When a miner has received more than 2/3 of votes. It will create a certificate for that block and broadcast it to other miners.
Step 3	Miners verify received certificates. If the miner node has received over 2/3 of valid certificates on the proposed block, the miner will add the block to the chain

In general, both the A2 and the proposed framework have the same security level when all transactions have to be verified by at least two-thirds of miners. In case the selected miner is honest, the verification time for transactions is not much different in both solutions. However, the proposed framework is better than the A2 in the case the selected miner is untrusted. More precisely, when the selected miner is compromised by hackers in a mining round, the hackers can broadcast a new block containing some invalid transactions. Of course, this new block will be removed by other miners in both solutions. However, for, A2, the valid transactions in the dropped block have to be reverified in the next mining round. Meanwhile, in the proposed framework, miners do not need to reverify these transactions because those transactions are still stored in VL until a block containing them has been successfully verified.

The proposed framework in this assumption can be used for consortium BC networks, in which untrusted miners must be less than one-third of the total number of miners in the networks.

CONCLUSION

This paper proposes a framework based on BC for the IoT. The proposed framework provides the reliability of data in the BC ledger and optimizes computational powers for miners in two assumptions about miners on the private BC network. With the assumption that miners are trusted nodes, the proposed framework operates more efficiently than other consensus protocols. In addition, in the case of existing untrusted miners with less than a third of the total number of miners, the proposed framework is better than the PBFT and Tendermint algorithms. The proposed framework is an effective and feasible solution for IoT networks.

ACKNOWLEDGMENTS

This research is funded by Vietnam National University of Ho Chi Minh City (VNU-HCMC) under grant number HTHT2021-18-02.