4134 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO.

6, JUNE 2020

An Efficient and Compacted DAG-Based

Blockchain Protocol for Industrial
Internet of Things
Laizhong Cui , Shu Yang , Ziteng Chen , Yi Pan , Mingwei Xu, Member, IEEE, and Ke Xu

Abstract—Industrial Internet of Things (IIoT) has been Internet, allowing IIoT users to control their devices remotely
widely used in many fields. Meanwhile, blockchain is con- and share their resources efficiently. Nowadays, the IIoT has
sidered promising to address the issues of the IIoT. How- been utilized in many fields, including manufacturing, energy
ever, the current blockchains have a limited throughput. In
this article, we devise an efficient and secure blockchain trading, healthcare, vehicles, etc. [1]. However, there are some
protocol compacted directed acyclic graph (CoDAG) based ongoing challenges in the IIoT. First, security is a major
on a compacted directed acyclic graph, where blocks are problem, where the IIoT system may be easily attacked, and
organized in levels and width. New-generated blocks in the data may be tampered by malicious third parties. In [2], the
CoDAG will be placed appropriately and point to those in the
authors pointed out that the current IIoT system is vulnerable to
previous level, making it a well-connected channel. Trans-
actions in the network will be confirmed in a deterministic resist the distributed-denial-of-service attack. Meanwhile, the
period, and the CoDAG keeps a simple data structure at traditional IIoT system relies on centralized servers to maintain
the same time. We also illustrate the attack strategies by the network, which is threatened by the weak connectivity and
adversary, and it is proved that our protocols are resistant to cracked security [3]. Second, people are concerned about the
these attacks. Furthermore, we design a CoDAG-based IIoT privacy, since their data are exposed to network operators and
architecture to improve the efficiency of the IIoT system.
Experimental results show that the CoDAG achieves 164× other users. Third, the traditional IIoT architecture relies on
Bitcoin’s throughput and 77× Ethererum’s throughput. centralized servers to handle the data computation and storage,
and it will hit a bottleneck as the number of IIoT users explodes.
Index Terms—Blockchain, compacted directed acyclic
graph (CoDAG), industrial Internet of Things (IIoT).
In these years, blockchain has been regarded as a promising
solution to the issues of IIoT [4]–[6]. Blockchain is a decentral-
I. INTRODUCTION ized ledger for distributed parties in the network. Different from
the traditional system, blockchain provides consensus among
ECENTLY, the industrial Internet of Things (IIoT)
R has been considered a promising technology and has
attracted lots of attention. With the development of network
distributed participants without any centralized authority. Trans-
actions in the network are organized in blocks, and they are
immutable and auditable. Distributed nodes share the ledger,
communication, the IIoT devices, including sensors, smart
making blockchain a reliable and trustworthy system in the
objects, etc., will communicate and interact with others via the
distributed environment. Nowadays, blockchain technology has
been employed in the IIoT, and many blockchain-based IIoT
Manuscript received May 31, 2019; revised July 11, 2019; accepted applications have emerged [7]–[12]. Generally, the blockchain-
July 18, 2019. Date of publication July 25, 2019; date of current
version February 28, 2020. This work was supported in part by the based IIoT system has lots of advantages, including the
National Key R&D Program of China under Grant 2018YFB1800302 following.
and Grant 2018YFB0803405, in part by the National Natural Science 1) Blockchain provides reliable services to guarantee the
Foundation of China under Grant No. 61772345, Grant 61625203, and
Grant 61832013, in part by the China National Funds for Distinguished security and robustness of the IIoT.
Young Scientists under Grant 61825204, in part by Beijing Outstand- 2) Blockchain protects the sensitive data in the IIoT, since
ing Young Scientist Project, and in part by the Tencent “Rhinoceros only the user with a matched private key gets access to
Birds”—Scientific Research Foundation for Young Teachers of Shenzhen
University. Paper TII-19-2220. (Corresponding author: Shu Yang.) the data.
L. Cui, S. Yang, and Z. Chen are with the College of Computer Science 3) Users are incentivized to maintain the IIoT network, since
and Software Engineering, Shenzhen University, Shenzhen 518060, blockchain rewards those who make contributions to the
China (e-mail:, [email protected]; [email protected]; 18102720
[email protected]). system.
Y. Pan is with the Department of Computer Science, Georgia State Bitcoin [13] was the first blockchain protocol, where blocks in
University, Atlanta, GA 30302 USA (e-mail:, [email protected]). the network are organized in a linear chain. The author proposed
M. Xu and K. Xu are with the Beijing National Research Center for
Information Science and Technology and the Department of Computer the proof-of-work (PoW) mechanism (also called Nakamoto
Science and Technology, Tsinghua University, Beijing 100084, China consensus) for distributed nodes to achieve consensus. More
(e-mail:, [email protected]; [email protected]). specifically, miners will compete to solve the cryptographical
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. puzzles, and the winners are allowed to propose new blocks and
Digital Object Identifier 10.1109/TII.2019.2931157 get the corresponding rewards. Distributed miners in Bitcoin
1551-3203 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4135

share the ledger, such that the transaction information cannot be and protocols to provide the fast consensus speed. For those
tampered by any third party. IIoT devices that have greater computing power, they will act
In a blockchain-based IIoT system, lots of IIoT devices, as miners to maintain the distributed ledger, and their rewards
transactions, and sensitive resources and data are emerging will be the greatest. For the lightweight devices that have better
every moment; thus, a faster transaction processing and bandwidth, they will act as gateways to maintain the encrypted
consensus speed among the IIoT devices should be provided. communication channels and make transactions with others.
Nevertheless, the current throughput of blockchain is quite And other nodes with limited computing power and bandwidth
low. For example, Bitcoin only achieves seven transactions only need to provide the transaction payments and receive the
per second (tps), and Ethererum [14] only achieves 15 tps corresponding computational results. In this architecture, we
[15], and they cannot meet the transaction demand of the IIoT. will make full use of the resources in the IIoT, and the distributed
To improve the throughput of blockchain, different solutions IIoT devices can achieve fast consensus by using the CoDAG
were proposed, including modification on the linear chain protocols.
structure (e.g., Bitcoin-NG [16] and GHOST [17]), consensus To evaluate its performance, we implement a CoDAG proto-
mechanism (e.g., Algorand [18]), etc. Among them, solutions type, which is divided into data layer, chain layer, RPC/Console
based on the directed acyclic graph (DAG) structure [19]–[23] layer, and decentralized application (Dapp) layer. Among them,
allow multiple blocks to append the network concurrently; the chain layer is the most critical module, which is responsible
thus, they achieve the balance between asynchronization and for maintaining the network communication, data structure,
consensus speed. For the traditional DAG-based schemes miner management, and consensus algorithms. We evaluate our
(e.g., IOTA [19]), however, with more blocks appending the implementation in a real network environment, and the experi-
network, new-generated transactions will not be confirmed in a mental results show that CoDAG scales in width and blocksize.
deterministic period, and its security will decline accordingly. When the blocksize is 2 MB and the width is 15, the throughput
Although researchers developed other DAG-based schemes of the CoDAG is 1151 tps, which is 164× Bitcoin’s throughput
(e.g., Byteball [20], Spectre [21], Phantom [22], and Conflux and 77× Ethererum’s throughput, respectively. Meanwhile, we
[23]), their data structures are complicated, which will cause study the effect of different values of level formation time (the
additional and excessive computation to maintain the ledger. required duration when a level is full), and simulation results
Meanwhile, we point out that the current blockchain-based tell us that due to limited computing power and link delay,
IIoT architecture is irrational and inefficient. In the current the throughput of the CoDAG will stop growing if the level
system, every IIoT device is acting as a miner and making formation time is too small. Furthermore, we investigate the
computational contribution to the network. However, due to the block utilization (the proportion of valid blocks in the network),
limited battery, bandwidth, storage, and computing power, the and it will converge to 90% with the development of the network.
IIoT devices cannot provide enough computational services for The main contributions of this article are as follows:
the blockchain maintenance. Besides, many resources of IIoT 1) We devise an efficient and secure blockchain protocol
devices are wasted in the computation process, which will lead CoDAG to meet the transaction demand of the IIoT.
to inefficiency of the IIoT system. Hence, a new architecture 2) We design a compacted DAG structure with level and
for the IIoT should be developed to make the blockchain-based width and propose algorithms to maintain the network.
IIoT system more efficient. 3) We illustrate two basic attack strategies and a “hybrid”
In this article, we design an efficient and compacted DAG- strategy in CoDAG and prove that our CoDAG is resistant
based blockchain protocol CoDAG. Unlike the previous DAG- to them.
based solutions, we devise a compacted DAG structure with 4) We propose a CoDAG-based IIoT architecture, where the
level and width. Blocks in the CoDAG are organized in levels, CoDAG is utilized as the backbone protocols and data
and each level has a fixed width, which means the maximum structure, and the IIoT devices are divided into different
value of new-generated blocks in a round. Blocks in the cur- roles.
rent level will point to those in the previous level, such that The rest of this article is organized as follows. Section II will
the CoDAG will develop into a well-connected channel. We introduce related work of IIoT and blockchain. Section III will
design algorithms and protocols to place the new-generated present the data structure of the CoDAG and the CoDAG-based
blocks at a proper level, making sure that the CoDAG has IIoT architecture. Then, the protocols of the CoDAG and
the consistency and liveness property of blockchain. Compared well-connected channel will be studied in Sections IV and V.
with previous DAG-based schemes, transactions in the CoDAG Section VI will describe the CoDAG implementation framework
will be confirmed in a deterministic period that relies on the in detail. Simulation results will be shown in Section VII.
puzzle difficulty. Meanwhile, the CoDAG keeps a simple data Finally, Section VIII will conclude this article.
structure by modifying the existing DAG structure slightly
without introducing any additional data structure. Moreover, II. RELATED WORK
we demonstrate two basic attack strategies by adversaries and a
“hybrid” attack strategy that combines the previous strategies, A. Blockchain-Based IIoT Applications
and it is proved that our CoDAG is resistant to them. Due to the philosophy of decentralization, blockchain pro-
Furthermore, we design a CoDAG-based IIoT architecture, vides secure and reliable services for the IIoT. In the blockchain-
where the IIoT devices are divided into miner, gateway, and based IIoT system, data are stored in the distributed ledger,
node, and the CoDAG is utilized as the backbone data structure and the data computation process does not rely on any

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
 4136 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 6, JUNE 2020

centralized server. Besides, data privacy is guaranteed in the the consensus system (e.g., lightning network [35] and pegged
blockchain-based IIoT system, making sure that the sensitive sidechain [36]); and 2) on-chain model that improves and
industrial data will not be stolen by competitors. Additionally, modifies the traditional linear chain structure (e.g., Bitcoin-NG
the IIoT devices sharing their resources (e.g., data storage [16], GHOST [17], and IOTA [19]). For the off-chain models,
and computing) will be rewarded, which will encourage IIoT the lightning network [35] improves the throughput of Bitcoin
users to contribute to the network. These years have witnessed by building the micropayment channels between nodes. And
the emergence of many blockchain-based IIoT applications in pegged sidechains [36] are allowed to communicate with the
academia, and blockchain has been deployed in many industrial main chain through a two-way peg with desired flexibility.
fields, including energy trading, smart factory, logistics, etc. Similarly, other off-chain models rely on the trustworthy third
[24]. In [8], [25], and [26], blockchain technology is applied in parties, and they compromise decentralization. For the on-chain
energy trading and enhances the trading security and privacy. In models, on the one hand, the linear-based solutions, including
[9], blockchain combines the healthcare, making the sensitive Bitcoin-NG [16] and GHOST [17], are developed to improve the
medical data immutable and traceable. In [27], the authors efficiency, fairness, and usability of Bitcoin. However, in [33],
proposed a decentralized IIoT architecture based on blockchain it is proved that the linear-based structure is constrained by the
and introduced a whitelist mechanism for a smart factory. Liu maximum network delay in order to guarantee the security of
et al. [28] developed a blockchain-based anonymous reputation blockchain.
system and enhanced the retail marketing. And Huang et al. On the other hand, the DAG-based solutions [19]–[23] enjoy a
[29] devised a credit-based PoW mechanism and provided good throughput performance. Unlike the linear-based models,
a data authority management for the blockchain-based IIoT DAG-based solutions allow multiple blocks to attach to the
system. tail of the DAG structure simultaneously. Among them, IOTA
However, we note that the current blockchain-based IIoT sys- [19] provides the fast consensus service for distributed nodes,
tem utilizes the traditional blockchain platforms (e.g., Bitcoin, where the new-generated blocks will confirm and point to two
smart contracts, etc.), and they cannot provide fast consensus invalidated blocks existed in the network. Meanwhile, there are
for distributed nodes in the IIoT system. Besides, the current no miners in IOTA; thus, the users can make transactions without
blockchain-based IIoT system is not efficient enough, since any service charge. However, the transaction confirmation time
lots of computation resources are wasted. Therefore, in this and security of IOTA are not guaranteed as more blocks join the
article, we will discuss the throughput limitation of traditional network. Besides, the blocks in IOTA is not full ordered, such
blockchains and design a new blockchain protocol CoDAG and that there may exist some conflicting transactions, which will
a new blockchain-based IIoT architecture. not be successfully validated due to the absence of liveness. To
solve the issues in IOTA, some improved DAG-based solutions
were developed. Byteball [20] maintains the chain by a portion
B. Throughput of Current Blockchain of reputable witnesses elected by others and establishes a full
In a distributed environment, in order to achieve consensus order between blocks. However, it compromises decentraliza-
among nodes, many protocols were proposed, which can be cat- tion, since the ledger is maintained by a finite number of
egorized into permissioned protocol, permissionless protocol, designated nodes. In the academic area, the DAGlabs proposed
and hybrid protocol. For permissioned protocols (e.g., PBFT two protocols called Spectre [21] and Phantom [22]. Spectre
[30], Paxos [31]), only a finite number of designated nodes [21] adopts a voting mechanism that every block will submit its
with permission are selected to maintain the ledger, and its vote to a pair of previous blocks and try to determine the block
consensus speed will be very fast. Nowadays, permissioned order according to the majority of votes. Although Spectre has
protocols are applied in Hyperledger fabric [32], which allows a high throughput and fast transaction confirmation speed, it
users to develop their Dapp and achieve fast consensus in the still cannot be extended to support the full linear order over
distributed system. However, in permissioned protocols, only all blocks. To solve the order issue of Spectre, the DAGlabs
the designated parties are allowed to join the network, which devised Phantom [22] to support the full order of blockchain
will lead to centralization. For permissionless protocols, the maintained by honest nodes. However, Phantom compromises
Nakamoto consensus [13] is the first blockchain protocol applied the transaction speed and confirmation time. Conflux [23]
in Bitcoin, and nodes can dynamically join or leave the network. introduces the parent edge and reference edge and maintains
The Nakamoto consensus has two properties of blockchain [33], the pivot chain to keep the full order of blockchain, but its
[34]: 1) consistency: distributed miners in the network have the confirmation time is still nondeterministic.
consistent view of the chain; and 2) liveness: a new-submitted Moreover, Pass and Shi designed the hybrid protocol [37]
transaction will appear in the ledger of all miners. However, the that combines the permissioned and permissionless protocols
throughput of Bitcoin is only 7 tps, which cannot keep up with and devised Thunderella [38] to support instant transactions.
the demand in the IIoT world. However, the performance of hybrid protocol is limited, since
Many improved permissionless protocols were designed to the permissionless protocol is still employed as the backbone
improve the throughput of blockchain, which can be categorized consensus in the hybrid protocol.
into the following: 1) off-chain model that builds additional As a summary, a comparison between different consensus
chains outside the main chain and reduces the participants in protocols is presented in Table I.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4137

TABLE I
COMPARISON BETWEEN DIFFERENT CONSENSUS PROTOCOLS

TABLE II
NOTATION LIST

Fig. 1. Simple example for CoDAG.

In the CoDAG, the leaf nodes of the graph are tips, and the
tips with largest connectivity are denoted as navigators. For
example, the navigators in Fig. 1 are nodes 13–15, since they
III. BASIC IDEAS OF THE CODAG share the value of connectivity. Furthermore, to maintain the
structure of the CoDAG, we define reverse connectivity Rv of
A. Data Structure of the CoDAG node v as the number of paths from navigators to it. In Fig. 1,
In this article, we devise a blockchain structure CoDAG, R8 = 9, including nine paths:
where the DAG structure is employed as the backbone data 13 → 10 → 8, 13 → 11 → 8, 13 → 12 → 8,
structure. The advantage of the DAG structure is that it fits 14 → 10 → 8, 14 → 11 → 8, 14 → 12 → 8,
well in asynchronous environment; thus, it can achieve high 15 → 10 → 8, 15 → 11 → 8, 15 → 12 → 8.
throughput among distributed nodes. Different from previous To guarantee the security of the CoDAG, the honest nodes
DAG-based blockchain schemes, the CoDAG is more like a in the network need to be connected closely and enjoy a higher
channel within fixed width, and blocks are closely connected reverse connectivity for the current navigator. The notations of
in adjacent levels. Therefore, as blocks join the network, the the CoDAG are listed in Table II.
CoDAG will always develop in a compacted DAG shape.
We call G = (V, E) the graph, where V is the set of blocks or
B. CoDAG-Based IIoT Architecture
nodes, and E is the set of edges between nodes. We denote lv as
the level of node v, which means the value of longest path from Generally, the IIoT devices are limited by their battery,
the genesis to it. wl is denoted as the width of level l, which is computing power, storage, communication bandwidth, etc. In
constrained by K, that is, ∀l, wl < K. The width also means the the IIoT network, due to the huge transaction number, the
maximum value of nodes that can concurrently join the network overwhelming transactions cannot be handled by IIoT devices,
in a round. It can also be adjusted or self-adaptive by the block and lots of resources are wasted in the computation process,
generation rate. which will lead to inefficiency of the system. To make full use
In the graph, nodes in current level will point to those in of the computation and other network resources, we design a
the previous level. To measure how closely a node connects framework shown in Fig. 2, where the IIoT devices are divided
with others, we define connectivity Cv ,w as the number of into miner, gateway, and node.
paths between node v and w. Specifically, the connectivity from 1) Miner: A miner will maintain the ledger using our
node v to itself is 1, that is, Cv ,v = 1. Besides, we define the CoDAG protocols. Similar to Bitcoin, the miners in our
connectivity of node v as the number of paths from genesis to it, architecture will use their computing power to bundle
which is denoted as Cv . In Fig. 1, C8 = 9, including nine paths: the transactions into blocks and append the blocks to
8 → 4 → 1 → 0, 8 → 4 → 2 → 0, 8 → 4 → 3 → 0, our proposed DAG-based structure using the CoDAG
8 → 5 → 1 → 0, 8 → 5 → 2 → 0, 8 → 5 → 3 → 0, protocols. Miners will get the greatest reward, since their
8 → 6 → 1 → 0, 8 → 6 → 2 → 0, 8 → 6 → 3 → 0. contribution to the network is the most.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
 4138 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 6, JUNE 2020

Fig. 2. CoDAG ledger framework.

2) Gateway: A gateway can be a lightweight IIoT device

with enough bandwidth resources, including 5G base
stations connecting other IIoT devices, etc. A gateway
is responsible for downloading the part of the ledger,
maintaining the encrypted communication channels with
other IIoT devices, and making transactions with other
gateways.
3) Node: A node will submit the payments and other data
to the gateway and receive the corresponding results. If a
device has enough computing power or bandwidth, it can
be a miner or gateway as well.
In this architecture, the CoDAG is utilized as the backbone allowed to bundle transactions into a block, and append the block
data structure and protocol, which can provide the fast consensus to the end of the linear chain, and get the corresponding rewards.
speed to meet the transaction demand in the IIoT world. Due However, due to the propagation delay, forks will happen when
to the profound throughput of the CoDAG, the miners in our a block is mined by multiple miners simultaneously. When the
CoDAG-based IIoT architecture can achieve a fast transaction forks exist, honest miners will append new blocks to the longest
processing speed; thus, the transaction fee for each payment chain.
will be relatively small, and the IIoT devices are more willing 2) Protocol of the CoDAG: Similar to the Nakamoto consen-
to make transactions in the network. Meanwhile, we make full sus, in the CoDAG, transactions will be enveloped into blocks by
use of the resources of IIoT devices and deploy the lightweight miners who successfully solve the cryptographical puzzles. And
nodes (e.g., the gateway) to help maintain the ledger, which will the miners who successfully mine a new block will get rewarded;
improve the efficiency of our CoDAG-based IIoT system. In the thus, miners are incentivized to contribute to the network. Due
future, to improve the efficiency further, we will introduce the to the network latency, forks may appear, where more than K
hybrid consensus to the CoDAG, and the system will be divided nodes join the same level. In this case, nodes with higher reverse
into fast mode and slow mode. In the slow mode, miners will connectivity will have a higher priority compared with other
run the PoW protocol and compete to join the membership of nodes who have a lower reverse connectivity, and the top K
the Byzantine agreement and enter the fast mode, and those who nodes in level l with higher reverse connectivity will be defined
successfully enter the fast mode will achieve the consensus and as candidates, and the candidate set is denoted as Cl . A special
maintain the ledger more rapidly and get their corresponding case is that several nodes with the same priority belong to the
rewards. borderline candidates, and they will be considered as candidates
and share the rewards.
IV. PROTOCOLS OF THE CODAG
B. Algorithms of the CoDAG
A. Overview
1) Block Generation Algorithm: To compute where a new
To clarify the protocol of the CoDAG, a brief introduction of block should be placed in the CoDAG, we design Generate()
the Nakamoto consensus will be given. as the main function (see Algorithm 2) and Connectivity() (see
1) Nakamoto Consensus in a Nutshell: The Nakamoto con- Algorithm 1) as the subfunction to compute the connectivity of a
sensus is applied in Bitcoin and provides the PoW mechanism node. The input of Generate() is the graph G and the new block
for distributed nodes to achieve consensus. Participants in the x, and the output is the new graph with x and related edges.
network, called miners, will try their best to solve a random In a graph G = (V, E), L represents the maximum level,
math puzzle. The miners who successfully solve the puzzle are and v, w denotes an edge pointing from node v to node w.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4139

First, from line 2 to line 5, Select() will call Connectivity()

to compute the connectivity of each node and choose the
navigators that has the largest values of connectivity. Line 7
to line 11 is the computation process of reverse connectivity.
Similarly, Algorithm 3 also adopts the traversal method level
by level iteratively. Contrary to Algorithm 1, the traverse order
of Select() is from the outer level L − 1 to the inner level,
since the reverse connectivity is the value of paths pointing
from the navigators. And the reserve connectivity of navigators
in maximum level L has the same values of 1. In Fig. 1, for
example, suppose we are computing the reverse connectivity
of node 8 in l3 . Since there are edges 10, 8, 11, 8, 12, 8
connecting the nodes between l3 and l4 , the algorithm will
add the reverse connectivity of nodes 10–12 to node 8, that is,
R8 = R8 + R10 , R8 = R8 + R11 , and R8 = R8 + R12 . Finally,
line 12 to line 19 is the candidate selection process. If there are no
more than K nodes in level l, they will be selected as candidates
of this level. Otherwise, Select() will sort and choose the top K
nodes that have the highest values of reserve connectivity as the
candidates of level l.

V. WELL-CONNECTED CHANNEL OF THE CODAG

Different from traditional linear chain, the CoDAG employs
a compacted DAG-based structure that is more suitable in asyn-
chronous environment, and it will develop into a well-connected
channel whose width can be adjusted flexibly. To attack the
network, adversaries will try to control the navigators, such
When a new-generated block x appears and wants to join the net- that the new-generated blocks will point to the adversarial
work, Generate() (see Algorithm 2) will call the Connectivity() navigators. Therefore, we consider a situation where an attacker
(see Algorithm 1) to compute the connectivity of every node builds an adversarial channel with higher connectivity and
in G. In Connectivity(), we adopt the traversal method level controls the navigators. Similar to Bitcoin [13], we have the
by level iteratively from line 5 to line 9. The principal of following notations:
Connectivity() is that, for every node v in level l, if there are 1) p = proportion of honest computing power;
directed edges connecting node v and nodes in the last level, 2) q = proportion of adversarial computing power;
the algorithm will add the connectivity of nodes in the last level 3) qz = probability of an adversary catching up from z levels
to node v. For example, in Fig. 1, suppose we are computing behind.
the connectivity of node 8 in l3 , and there are nodes 4–6 in l2 . Lemma 1:
We note that the directed edges < 8, 4 > connects nodes 4 and 
1, if p ≤ q.
8; thus, the algorithm will add the connectivity of nodes 4–8, qz = . (1)
that is, C8 = C8 + C4 . Similarly, Connectivity() will execute (q/p)z , otherwise
C8 = C8 + C5 and C8 = C8 + C6 in l2 , and we will get the
connectivity C8 of node 8. Connectivity() will start from the If honest and adversarial nodes start to mine the next w
first level l0 , compute the connectivity of other nodes in the blocks simultaneously, the winning rate of honest party is p,
following levels iteratively, and, finally, return the results to and adversary party is q. Hence, the proof of Lemma 1 is the
Generate(). same as [13].
In Generate(), the node x in level L will be placed according Then, we investigate the transaction confirmation time.
to the connectivity and point to K nodes that has the largest Lemma 2: If node v is z levels away from largest level L,
values of connectivity in level L − 1. Meanwhile, if level L the possibility P of deprivation of its candidate identity is:
is full of K nodes, the value of L will increase by 1. Finally, z
  z −k 
the network information G = (V, E) will be updated in lines 6  λk e−λ q q
P =1− 1− , whereλ = z .
and 7. k! p p
k =0
2) Candidate Selection Algorithm: To deal with forks in the
CoDAG, we develop Select() (see Algorithm 3) to select the Proof: We suppose there are two nodes in level l, where node
candidates of a level according to their reverse connectivity. a is produced by honest miners and b is produced by adversary
The input of Algorithm 3 is the graph G = (V, E) and level l, miners, and suppose Ra > Rb when the largest level of the
and the output is the candidate set in l. graph L satisfies L − l ≤ z.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
 4140 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 6, JUNE 2020

We first prove that in any level k > l, for the two honest nodes
vi and vj , then Cv i ,a = Cv j ,a and Cv i ,b = Cv j ,b ; for the two
adversary nodes wi and wj , Cw i ,a = Cw j ,a and Cw i ,b = Cw j ,b .
We prove it through induction. In level l + 1, all honest nodes
will point to a and avoid pointing to b; thus, their connectivity
to a is 1 and connectivity to b is 0; all adversary nodes will
point to b and avoid pointing to a; thus, their connectivity to
a is 0, and connectivity to b is 1. For a level k > l, suppose
the nodes in the level satisfy the assumption, and there are
P honest nodes and Q adversary nodes, any honest node v k
satisfies Cv k ,a = X and Cv k ,b = Y , and any adversary node
wk satisfies Cw k ,a = X  and Cw k ,b = Y  . Then, in level k + 1,
for any honest node v k +1 , Cv k + 1 ,a = X ∗ P + X  ∗ (K − P )
and Cv k + 1 ,b = Y ∗ P + Y  ∗ (K − P ); for any adversary node
wk +1 , Cw k + 1 ,a = X  ∗ Q + X ∗ (K − Q) and Cw k + 1 ,b = Y  ∗
Q + Y ∗ (K − Q).
Then, we prove the above Lemma. For a level k > l, R(k) =
∀v k Rv k is the totalreverse connectivity of all honest nodes in
level k, and R̂(k) = ∀v k Rw k is the total reverse connectivity
of all adversary nodes. Suppose R̂(k )
R(k ) ≥  should be satisfied
to make the adversary party catch up with the honest party. In
level k + 1, we will have R(k) = R(k + 1) ∗ P + R̂(k + 1) ∗
(K − Q) and R̂(k) = R̂(k + 1) ∗ Q + R(k + 1) ∗ (K − P )
indicating that all honest nodes will point to honest nodes first,
and all adversary nodes will point to adversary nodes first. Based
on the above equations, we can infer that R̂(k +1) ∗P +P −K
R(k +1) ≥ Q +Q ∗−∗K .
Because P and Q follow two independent Poisson distribu-
tions, the expectation value E( R̂(k +1) ∗E (P )+E (P )−K
R(k +1) ) ≥ E (Q )+E (Q )∗−∗K ≥
 ∗ E( QP
) =  ∗ pq , where E( QP
) is the ratio of the computing
power of the honest party to the computing power of the
adversary party.  Fig. 3. Two attack strategies in the CoDAG. (a) Adversary builds an
alternate channel. (b) Adversary competes in the original channel.
An example. Suppose there is an adversary trying to conduct
a malicious behavior (e.g., double-spending attack [39], [40]).
Similar to Bitcoin, the attacker needs to append the blocks TABLE III
that contains his own forged transaction records to the network POSSIBILITY OF CANDIDATE IDENTITY DEPRIVATION
and control the candidate identity of the CoDAG, such that the
following new-generated blocks will verify and point to those
adversary candidates.
We will compute the possibility of the candidate identity
deprivation of node v in different values of q and z (shown in
Table III).
We can see that given a fixed value of q, when the value
of z grows, the possibility of the candidate identity deprivation
will decrease exponentially. Therefore, as the network develops,
our CoDAG protocol is secure to resist the candidate identity
deprivation.
We point out that there are two attack strategies that may
happen in the CoDAG. The first is that the adversary will build will be given to honest nodes. The strategy for an honest node
a completely separate new channel [see Fig. 3(a)]. In this case, is just opposite.
the adversary will point to adversarial nodes in the new channel In the second strategy, each node has K votes. For the honest
rather than the honest node in the original channel, which is the nodes, they will first give their votes to honest nodes and then
same as Bitcoin. randomly give their remaining votes to adversary nodes in
The second is that the adversaries will compete in the original previous level. Contrary to honest party, the adversary nodes will
channel with honest nodes [see Fig. 3(b)]. For an adversary, it first give their votes to adversary nodes and then randomly give
will first point to adversarial nodes, and the remaining pointers their remaining votes to honest nodes. However, in Lemma 2,

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4141

Fig. 4. Implementation framework.

it is proven that, if the honest party dominates the computing

power, they will get most votes. Meanwhile, as more nodes
join the network and number of levels grows, honest votes will
get accumulated, and the security of the CoDAG will increase
further. Hence, the CoDAG is secure enough to resist these two
attack strategies.
We note that for the same level, if there are 2 · K nodes, then
the attack strategy belongs to the first one, since the adversary
will build an alternate channel whose width is also K. And if
there are less than 2 · K nodes, the attack strategy belongs to
the second one. Moreover, we point out that the adversary will
conduct a “hybrid” attack strategy, where the first strategy is
used in some levels, and the second strategy is used in other B. Implementation Details
levels. However, our CoDAG protocol is still resistant to the
“hybrid” strategy, which can be proved based on the proofs of The communication and synchronization process is shown
two basic strategies we mentioned, and the details are omitted in Algorithm 4. From line 2 to line 4, nodes will build the
in this article. Therefore, these two basic attack strategies can connections with others by handshake(peer) and choose the
cover the “hybrid” strategy by adversary, and the security is best peers in good condition. In lines 6 and 7, peer nodes means
guaranteed in our CoDAG protocols. the navigators in peer info, and dag nodes means the navigators
in peer. From line 8 to line 9, if any difference is found in the
VI. IMPLEMENTATION current level, CoDAG will merge the information of dag nodes
and peer nodes and synchronize the block information in this
A. Implementation Framework level. Then, in lines 10 and 11, peer nodes and dag nodes will
Fig. 4 demonstrates the framework of our implementation, point to blocks in previous level. The loop from line 8 to line
where the data layer, chain layer, RPC/Console layer, and Dapp 11 will not stop until peer nodes and dag nodes share the block
layer are involved. Among them, chain layer is the core layer in information at a certain level.
the CoDAG, which can be divided into the following. The process of inserting a block is given in Algorithm 5. Like
1) Network module: Synchronize the network information Bitcoin, transactions in the CoDAG are considered valid after
by peer-to-peer protocols. being confirmed in the last constant levels. From line 2 to line
2) CoDAG module: Maintain the data structure and 5, for ∀l where lv − l < constant, the algorithm will validate
algorithms. node and its containing transactions in peer node pool[l]. If
3) Miner module: Manage the miners and puzzle difficulty. node is valid, it will join the node pool. In line 6, node pool will
4) Consensus algorithm module: Maintain the consensus be sorted by their connectivity, and the top K nodes with the
among nodes. largest connectivity will become the navigators.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
 4142 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 6, JUNE 2020

TABLE IV
DEFAULT SIMULATION PARAMETERS

Algorithm 6 shows the process of updating the world state.

We define world state as the states of participants in the net-
work, where their account balances and transaction records are
involved. And we call stable level the level where the algorithm
stops visiting. We note that some jitters may happen at the
beginning stage, thus CoDAG will ignore the first stable_level Fig. 5. Network topology.
levels. From line 2 to line 9, ∀l where lv > stable level, existing
transactions in elected_node_pool will be visited and added to
tx_set. In line 6, the algorithm will sort the transactions in tx_set
by their amounts, making sure those transactions with a larger
amount can be bundled fast. Finally, CoDAG will execute the
tx_set and update the world_state of the network.

VII. EVALUATION
A. Experiment Setup
To evaluate CoDAG’s performance in a real environment,
computing devices with different computing power are de-
ployed, including a server, 5 Mac Minis, and 4 Lenovo laptops,
and they are connected in a LAN-based environment. In a
LAN-based environment, the link delay is very small; thus, the
Fig. 6. Simulation equipment.
information propagation delay of the CoDAG will be negligible.
To emulate a real-world network environment, we manually add
some link propagation delay to the network. More specifically, required duration when a level is full of K blocks. Assume
we use the NetEm [41], which is a network emulation module the level formation time of the system is T ; then, the average
T
to simulate the link delay in a LAN environment. In this article, proposal time of every single block will be K . Therefore, the
the average delay time is set to be 50 ms [42]. In this case, the level formation time will directly influence the block proposal
link latency will be added to the network; thus, we can evaluate time in the network and then influence the throughput of the
our CoDAG prototype in a real network environment. CoDAG. In this article, we will study the relationship between
Additionally, we design a script to produce enough transac- the throughput and different values of level formation time. We
tions with random amounts and addresses. In the CoDAG, every will change the level formation time by altering the mining
miner has a transaction pool to store the transactions that send difficulty of the system. The more mining difficulty is, the more
to it. Initially, we let the trigger node run the script, and the level formation time it will be.
new-generated transactions will be stored in the matched miner Besides, we denote the block utilization as the proportion of
pools according to the transaction addresses and wait to be valid blocks in the network. In our prototype, if the width of a
bundled by the corresponding miners. In this case, miners will level exceeds K, some blocks will become invalid and wait to
try their best to bundle transactions in their pools and append reappend to the network again. Hence, we use block utilization
the block to the network. to measure the percentage of valid blocks of the system,
In our experiments, we will evaluate the CoDAG’s perfor- and investigate the relationship between block utilization and
mance in different values of blocksize and width. The blocksize different values of blocksize and width. The default simulation
will scale from 10 kB to 2 MB, and the width will vary from parameters are listed in Table IV. The simulation topology and
1 to 20. Meanwhile, we define level formation time as the equipment are presented in Fig. 5 and 6 respectively.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4143

Fig. 7. Relationship between throughput and width. Fig. 9. Throughput of the CoDAG in different level formation time when
blocksize = 20 kB.

Fig. 8. Relationship between throughput and blocksize.

Fig. 10. Throughput of the CoDAG in different level formation time
when blocksize = 30 kB.

B. Experiment Results
1) Throughput in Different Blocksizes and Widths: Figs. 7 larger values of width and blocksize outperforms the smaller
and 8 present the CoDAG’s throughput in different blocksizes one. Especially, when the blocksize is 2 MB and the width is
and widths. In Fig. 7, we can observe that given a fixed blocksize, 15, the CoDAG achieves the highest throughput of 1151 tps,
as the width grows, the throughput of CoDAG shows an in- which is 164× Bitcoin’s and 77× Ethereum’s. Moreover, we
creasing trend. Meanwhile, given a fixed width, as the blocksize notice that the throughput of the CoDAG soars as the blocksize
grows, the throughput will rise accordingly. For example, when grows from 32 to 256 kB. However, when the blocksize reaches
the blocksize is fixed at 30 kB, the CoDAG with a width of a larger value, the throughput presents a slower growth. The
20 can achieve 394 tps, which is larger than other widths. reason is that, due to the propagation latency, a block with a
Moreover, when the width is fixed at 20, the CoDAG with a larger blocksize will contain excessive transactions that cannot
blocksize of 30 kB is 2.6 times as blocksize = 10 kB and 1.9 be handled in time. Additionally, limited computing power of
times as blocksize = 20 kB, respectively. Besides, the CoDAG miners is also one of the restrictions.
always performs better than Bitcoin and Ethererum. In Fig. 7, 2) Throughput in Different Level Formation Time: Figs. 9
the CoDAG can achieve 394 tps, which is 56× Bitcoin’s and and 10 depict the throughput in different level formation time.
26× Ethererum’s. Meanwhile, we also notice some throughput Generally, we can observe that the CoDAG with a longer level
fluctuations as the width of the CoDAG grows. If blocksize formation time performs worse than smaller ones. In Fig. 9,
is fixed at 30 kB, when the width increases from 13 to 14, when the blocksize is fixed at 20 kB, the CoDAG with 18 s
the throughput of the CoDAG presents a slight decline. The performs worse than 8 and 13 s. Moreover, we notice no
reasons are twofolds: 1) a larger width will bring more block obvious deviation between 8 and 13 s, which means when level
information for distributed nodes to synchronize and lead to formation time declines to a relatively small value, it will have
longer propagation delay; and 2) there is some instability of the a limited effect on CoDAG’s throughput. In theory, less level
transaction generation rate, and the amount of transactions may formation time will lead to faster block generation rate, and the
be instable, and one of our future works is to improve it. throughput will rise accordingly. However, due to the capacity,
Fig. 8 also observes the relationship between throughput in excessive transactions and blocks cannot be computed and ver-
different blocksizes and widths. We can see that the CoDAG with ified by miners in time; thus, the throughput gap is not obvious.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
 4144 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 16, NO. 6, JUNE 2020

[5] Y. Xu, J. Ren, G. Wang, C. Zhang, J. Yang, and Y. Zhang, “A blockchain-

based nonrepudiation network computing service scheme for industrial
IoT,” IEEE Trans. Ind. Inform., vol. 15, no. 6, pp. 3632–3641, Jun. 2019.
[6] K. Huang et al., “Building redactable consortium blockchain for industrial
Internet-of-Things,” IEEE Trans. Ind. Inform., vol. 15, no. 6, pp. 3670–
3679, Jun. 2019.
[7] J. Kang, R. Yu, X. Huang, S. Maharjan, Y. Zhang, and E. Hossain,
“Enabling localized peer-to-peer electricity trading among plug-in hybrid
electric vehicles using consortium blockchains,” IEEE Trans. Ind. Inform.,
vol. 13, no. 6, pp. 3154–3164, Dec. 2017.
[8] X. Wu, B. Duan, Y. Yan, and Y. Zhong, “M2M blockchain: The case of
demand side management of smart grid,” in Proc. IEEE 23rd Int. Conf.
Parallel Distrib. Syst., Dec. 2017, pp. 810–813.
[9] M. Mettler, “Blockchain technology in healthcare: The revolution starts
here,” in Proc. IEEE 18th Int. Conf. e-Health Netw., Appl. Serv., Sep. 2016,
pp. 1–3.
[10] H. Liu, Y. Zhang, and T. Yang, “Blockchain-enabled security in electric
Fig. 11. Block utilization of the CoDAG. vehicles cloud and edge computing,” IEEE Netw., vol. 32, no. 3, pp. 78–83,
May/Jun. 2018.
[11] Z. Su, Y. Wang, Q. Xu, M. Fei, Y. Tian, and N. Zhang, “A secure
charging scheme for electric vehicles with smart communities in energy
When the blocksize is fixed at 30 kB, similar results can be blockchain,” IEEE Internet Things J., vol. 6, no. 3, pp. 4601–4613,
Jun. 2019.
obtained, as depicted in Fig. 10. [12] W. Liang, M. Tang, J. Long, X. Peng, J. Xu, and K. Li, “A secure fabric
3) Block Utilization: Fig. 11 provides the block utilization in blockchain-based data transmission technique for industrial internet-of-
different blocksizes and widths. Generally, the block utilization things,” IEEE Trans. Ind. Inform., vol. 15, no. 6, pp. 3582–3592, Jun. 2019.
[13] S. Nakamoto et al., “Bitcoin: A peer-to-peer electronic cash system,” 2008.
will converge to 90% as width grows. This is because, if the [14] G. Wood et al., “Ethereum: A secure decentralised generalised transaction
width of the CoDAG is too small, the new-generated block will ledger,” Ethereum Project Yellow Paper, vol. 151, pp. 1–32, 2014.
not be aware that the target level has been filled with K blocks [15] L. Bach, B. Mihaljevic, and M. Zagar, “Comparative analysis of
blockchain consensus algorithms,” in Proc. 41st Int. Conv. Inf. Commun.
already. In this case, the new block will exceed the maximum Technol., Electron. Microelectron., 2018, pp. 1545–1550.
value of width and be regarded an invalid block; thus, the block [16] I. Eyal, A. E. Gencer, E. G. Sirer, and R. Van Renesse, “Bitcoin-NG: A
utilization is low. As the width increases, the percentage of scalable blockchain protocol,” in Proc. 13th USENIX Symp. Netw. Syst.
Des. Implementation, 2016, pp. 45–59.
invalid blocks will decrease. [17] Y. Sompolinsky and A. Zohar, “Secure high-rate transaction processing
in bitcoin,” in Proc. Int. Conf. Financial Cryptography Data Secur., 2015,
VIII. CONCLUSION pp. 507–527.
[18] Y. Gilad, R. Hemo, S. Micali, G. Vlachos, and N. Zeldovich, “Algorand:
In this article, we devised an efficient and compacted DAG- Scaling byzantine agreements for cryptocurrencies,” in Proc. 26th Symp.
Oper. Syst. Princ., 2017, pp. 51–68.
based blockchain protocol CoDAG. Unlike the previous DAG- [19] S. Popov, “The tangle,” 2016. [Online]. Available: https://iota.org/
based solutions, blocks in the CoDAG were organized in a DAG IOTA_Whitepaper.pdf
structure with a fixed width. We developed protocols and algo- [20] A. Churyumov, “Byteball: A decentralized system for storage and transfer
of value,” 2016. [Online]. Available: https://byteball.org/Byteball.pdf
rithms to maintain and secure the network, and transactions will [21] Y. Sompolinsky, Y. Lewenberg, and A. Zohar, “Spectre: Serialization of
be confirmed in a deterministic period. We presented possible proof-of-work events: Confirming transactions via recursive elections,”
attack strategies in the CoDAG and proved that it can resist these 2016. [Online]. Available: https://eprint.iacr.org/2016/1159.pdf
[22] Y. Sompolinsky and A. Zohar, “Phantom: A scalable blockdag protocol,”
attacks. We also designed a CoDAG-based IIoT architecture 2018. [Online]. Available: https://eprint.iacr.org/2018/104.pdf
to achieve a better performance for the IIoT system. Finally, [23] C. Li, P. Li, W. Xu, F. Long, and A. C.-C. Yao, “Scaling Nakamoto con-
we implemented a CoDAG prototype, and simulation results sensus to thousands of transactions per second,” 2018, arXiv:1805.03870.
[24] M. S. Ali, M. Vecchio, M. Pincheira, K. Dolui, F. Antonelli, and
showed that CoDAG outperforms the Bitcoin and Ethererum, M. H. Rehmani, “Applications of blockchains in the internet of things:
which is suitable and efficient for the IIoT. A comprehensive survey,” IEEE Commun. Surv. Tut., vol. 21, no. 2,
In the future, to improve the CoDAG further, some modi- pp. 1676–1717, 2nd Q 2019.
[25] N. Z. Aitzhan and D. Svetinovic, “Security and privacy in decentralized
fication schemes will be proposed, including 1) tuning-up the energy trading through multi-signatures, blockchain and anonymous mes-
parameters (e.g., blocksize and width), and 2) combining with saging streams,” IEEE Trans. Dependable Secure Comput., vol. 15, no. 5,
the hybrid consensus protocols. pp. 840–852, Sep./Oct. 2016.
[26] Z. Li, J. Kang, R. Yu, D. Ye, Q. Deng, and Y. Zhang, “Consortium
blockchain for secure energy trading in industrial Internet of Things,”
REFERENCES IEEE Trans. Ind. Inform., vol. 14, no. 8, pp. 3690–3700, Aug. 2017.
[27] J. Wan, J. Li, M. Imran, and D. Li, “A blockchain-based solution for
[1] L. Da Xu, W. He, and S. Li, “Internet of Things in industries: A survey,” enhancing security and privacy in smart factory,” IEEE Trans. Ind. Inform.,
IEEE Trans. Ind. Inform., vol. 10, no. 4, pp. 2233–2243, Nov. 2014. vol. 15, no. 6, pp. 3652–3660, Jun. 2019.
[2] Q. Yan, W. Huang, X. Luo, Q. Gong, and F. R. Yu, “A multi-level [28] D. Liu, A. Alahmadi, J. Ni, X. Lin, and X. Shen, “Anonymous reputation
DDOS mitigation framework for the industrial Internet of Things,” IEEE system for IIoT-enabled retail marketing atop POS blockchain,” IEEE
Commun. Mag., vol. 56, no. 2, pp. 30–36, Feb. 2018. Trans. Ind. Inform., vol. 15, no. 6, pp. 3527–3537, Jun. 2019.
[3] X. Sun and N. Ansari, “Dynamic resource caching in the IoT application [29] J. Huang, L. Kong, G. Chen, M. Wu, X. Liu, and P. Zeng, “Towards
layer for smart cities,” IEEE Internet Things J., vol. 5, no. 2, pp. 606–613, secure industrial IoT: Blockchain system with credit-based consensus
Apr. 2018. mechanism,” IEEE Trans. Ind. Inform., vol. 15, no. 6, pp. 3680–3689,
[4] N. Teslya and I. Ryabchikov, “Blockchain-based platform architecture Jun. 2019.
for industrial IoT,” in Proc. 21st Conf. Open Innov. Assoc., Nov. 2017, [30] M. Castro and B. Liskov, “Practical byzantine fault tolerance and proactive
pp. 321–329. recovery,” ACM Trans. Comput. Syst., vol. 20, no. 4, pp. 398–461, 2002.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.
CUI et al.: EFFICIENT AND COMPACTED DAG-BASED BLOCKCHAIN PROTOCOL FOR INDUSTRIAL INTERNET OF THINGS 4145

[31] L. Lamport et al., “Paxos made simple,” ACM Sigact News, vol. 32, no. 4, Yi Pan received the B.E. and M.E. degrees
pp. 18–25, 2001. in computer engineering from Tsinghua Univer-
[32] E. Androulaki et al., “Hyperledger fabric: A distributed operating system sity, Beijing, China, in 1982 and 1984, respec-
for permissioned blockchains,” in Proc. 13th EuroSys Conf., 2018, p. 30. tively, and the Ph.D. degree in computer science
[33] R. Pass, L. Seeman, and A. Shelat, “Analysis of the blockchain protocol from the University of Pittsburgh, Pittsburgh, PA,
in asynchronous networks,” in Proc. Annu. Int. Conf. Theory Appl. USA, in 1991.
Cryptograph. Techn., 2017, pp. 643–673. He is currently a Regents’ Professor and
[34] J. Garay, A. Kiayias, and N. Leonardos, “The bitcoin backbone protocol: Chair of Computer Science with Georgia State
Analysis and applications,” in Proc. Annu. Int. Conf. Theory Appl. University, Atlanta, GA, USA. He has served
Cryptograph. Techn., 2015, pp. 281–310. as an Associate Dean and Chair of Biology
[35] J. Poon and T. Dryja, “The bitcoin lightning network: Scalable off-chain Department from 2013 to 2017 and the Chair
instant payments,” 2016. of Computer Science from 2006 to 2013. His profile has been featured
[36] A. Back et al., “Enabling blockchain innovations with pegged sidechains,” as a distinguished alumnus in both Tsinghua Alumni Newsletter and
2014. [Online]. Available: https://blockstream.com/sidechains.pdf University of Pittsburgh CS Alumni Newsletter. He has authored or coau-
[37] R. Pass and E. Shi, “Hybrid consensus: Efficient consensus in the thored more than 250 journal papers with more than 90 papers published
permissionless model,” in Proc. 31st Int. Symp. Distrib. Comput., 2017, in various IEEE journals. In addition, he has authored or coauthored
pp. 39:1–39:16. 194 papers in refereed conferences. He has also coauthored/coedited
[38] R. Pass and E. Shi, “Thunderella: Blockchains with optimistic instant 44 books. His work has been cited more than 10 000 times in Google
confirmation,” in Proc. Proc. Annu. Int. Conf. Theory Appl. Cryptograph. Scholar. His current H-index is 54. He has served as an Editor-in-Chief
Techn., 2018, pp. 3–33. or editorial board member for 20 journals including seven IEEE Transac-
[39] G. O. Karame, E. Androulaki, and S. Capkun, “Double-spending fast tions. He has organized many international conferences and delivered
payments in bitcoin,” in Proc. ACM Conf. Comput. Commun. Secur., 2012, keynote speeches at more than 60 international conferences around the
pp. 906–917. world. His research interests include parallel and cloud computing, big
[40] M. Rosenfeld, “Analysis of hashrate-based double spending,” 2014, data, and bioinformatics.
arXiv:1402.2009. Dr. Pan is the recipient of many awards including an IEEE Trans-
[41] S. Hemminger et al., “Network emulation with NetEm,” in Proc. Linux actions Best Paper Award, IEEE Conference Best Paper Awards, and
Conf. AU, 2005, pp. 18–23. several other conference and journal Best Paper Awards, four IBM
[42] L. T. Nguyen, R. Harris, and J. Jusak, “Analysis of networking and Faculty Awards, two Japan Society for the Promotion of Science Senior
application layer derived metrics for web quality of experience,” in Proc. Invitation Fellowships, the IEEE Outstanding Leadership Award, the
IEEE Consum. Commun. Netw. Conf., 2012, pp. 321–325. IEEE International Conference on BioInformatics and BioEngineering
(BIBE) Outstanding Achievement Award, the National Science Founda-
tion (NSF) Research Opportunity Award, and the Air Force Office of
Laizhong Cui received the B.S. degree in com- Scientific Research (AFOSR) Summer Faculty Research Fellowship.
puter science and technology from Jilin Uni-
versity, Changchun, China, in 2007, and the
Ph.D. degree in computer science and technol-
ogy from Tsinghua University, Beijing, China, in
2012.
He is currently an Associate Professor with
the College of Computer Science and Software
Engineering, Shenzhen University, Shenzhen,
China. He led the projects of the National Key
Research and Development Program of China,
National Natural Science Foundation, and several projects of Guang-
dong Province and Shenzhen City. His research interests include future Mingwei Xu (M’08) received the B.Sc. and Ph.D.
Internet architecture, edge computing, big data, Internet of Things, degrees in computer science and technology
computational intelligence, software-defined networks, and machine from Tsinghua University, Beijing, China.
learning. He is currently a Full Professor with the De-
partment of Computer Science, Tsinghua Uni-
versity. His research interests include computer
network architecture and high-speed router
Shu Yang received the B.Sc. degree from the architecture.
Beijing University of Posts and Telecommunica-
tions, Beijing, China, in 2009, and the Ph.D. de-
gree from Tsinghua University, Beijing, in 2014,
both in computer science and technology.
His research interests include network
architecture and high-performance routers.

Ziteng Chen received the B.Sc. degree in com- Ke Xu received the B.Sc., M.Sc., and Ph.D. de-
munication engineering in 2018 from Shen- grees in computer science and technology from
zhen University, Shenzhen, China, where he is Tsinghua University, Beijing, China, in 1996,
currently working toward the M.Sc. degree in 1998, and 2001 respectively.
computer science and technology. He is with the Department of Computer
His research interests include software- Science and Technology, Tsinghua University,
defined networks and blockchain. Beijing. He is also with the Beijing National
Research Center for Information Science and
Technology, Beijing.

Authorized licensed use limited to: UNIVERSIDADE ESTADUAL DO CEARA. Downloaded on December 14,2021 at 14:16:31 UTC from IEEE Xplore. Restrictions apply.