1 BlockNDN A Bitcoin Blockchain Decentralized System Over Named Data Networking

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

BlockNDN:A Bitcoin Blockchain Decentralized

System over Named Data Networking

Tong Jin, Xiang Zhang, Yirui Liu, Kai Lei*


Institute of Big Data Technologies
Shenzhen Key Lab for Cloud Computing Technology & Applications
School of Electronic and Computer Engineering (SECE)
Peking University, SHENZHEN 518055, PR. CHINA
Email: [email protected], Corresponding Author*: [email protected]

Abstract Blockchain provides a new approach for to communicate with each other. They increase the possibility of
participants to maintain reliable databases in untrusted networks the existence of supernodes and bring some insecurity factors
without centralized authorities. However, there are still many such as frauds and single point of failure. Also, the TCP/IP
serious problems in real blockchain systems in IP network such as protocol does not provide native support for multicasting,
the lack of support for multicast and the hierarchies of status. In resulting in an increase in data transmission overhead.
this paper, we design a bitcoin-like blockchain system named
BlockNDN over Named Data Networking and we implement and Named Data Networking (NDN) project [5] is an important
deploy it on our cluster as well. The resulting design solves those candidate for next-generation internet architectures. NDN
problems in IP network. It provides completely decentralized network puts contents into first place. Contents are named
systems and simplifies system architecture. It also improves the following hierarchical naming rules and these names replace the
weak-connectivity phenomenon and decreases the broadcast role of IP addresses. Interest packets and data packets are the
overhead. two fundamental packet types in NDN. Each interest packet
corresponds to only one data packet. Nodes in NDN pull data
Keywords blockchain; NDN; peer-to-peer; bitcoin; next from networks using interest packets. The one side requests data,
generation networks the consumer, sends out an interest packet carrying the data
I. INTRODUCTION name and this interest will be forwarded to the other sides, the
producers. Producers then send back a data packet to the
Blockchain is a shared, unchangeable ledger for recording requester along the same route form where its interest has come.
the history of transactions [1]. It was first used in bitcoin, The intermediate routers with caches selectively store data
functioning as a data structure. Now it is abstracted from bitcoin packets forwarded before so that data can be sent back
system and has developed into a promising technology solution immediately once new consumers request for the same content.
which provides secure and valid achievement of distributed
consensus in areas such as supply chain & logistics, finance and good fit to the large amounts of static content via various P2P
healthcare. A blockchain is comprised of digitally recorded data overlay networks [6].
in packages called blocks. Each block contains the hash value of
the previous-block which comes out first and eventually forms a Does blockchain fit better on NDN than on IP? What are the
linear chain of blocks with the others. Different distributed advantages? To investigate on these questions, we implement a
storage solutions and consensus schemes should be designed in bitcoin-like blockchain system that runs on NDN network. In
a blockchain system depending on different requirements, which our system, the hash values function as the unique identification
enables nodes confirm and record information without of blocks. A node sends out an interest with the hash of the latest
authorities. To take bitcoin [2] as an example, it uses proof-of- block it stores locally. If the other node receives this interest,
work (PoW) [3] as consensus algorithm and designs full node finding that it has the same block in its blockchain and that block
storage that the so-called full nodes in the system should is not the latest one, it will collect all blocks behind into a data
maintain the whole blocks by themselves. packet to send to the consumer. In this way, the new joiner or
disconnected nodes can update their state quickly, that is, they
However, the growth of blockchain still hits bottlenecks and can get the whole blocks in the system in a short time. Nodes
impediments at present. The weak-connectivity and improper can also request specific blocks in the system using broadcast
protocols cause the propagation delay in IP network which mechanism. Our analysis shows that the bitcoin blockchain
causes blockchain forks in some systems [4]. Besides, although system over NDN circumvents those problems which may occur
blockchain is a fully decentralized system by definition, it is hard in IP network. The result shows that we have designed a more
to establish the connections between homogeneous nodes in decentralized and simpler system. Nodes do not need to get
practice. For example, with neighbor discovery in bitcoin, some information about other nodes like the identities and locations.
IP addresses of capable volunteers are recommended as the They just search for specific blocks data in the networks. Also,
connection options for users. These volunteers are called DNS NDN supports multicast and broadcast by nature, which reduced
seed nodes which function as infrastructure to help normal users the cost of message passing.Note that a NDN router with content

978-1-5090-4749-9/17/$31.00 2017 IEEE 75 ICUFN 2017


A blockchain with a linked list of block.
The core NDN packet dorwarding engine.
and Pending Interest Table (PIT). To request data, a data
store can work as a cache to optimize the system performance. consumer sends out an interest packet which carries the unique
name of the desired data. The intermediate NDN routers add an
The main contribution of this paper is summarized as below:
entry to the PIT of its own for recording the incoming interface
We creatively implement a blockchain system which is a where the interest packet comes in. Then they forward this
groundbreaking experiment in NDN and make our own packet by looking up the name in its FIB which has been
contribution to the promotion of NDN project. populated by name-based routing protocols rather than ip-
address-based [7]. Once the interest packet reaches a node with
We summarize the shortcomings of blockchain systems the requested back a data packet.
over IP networks which include the possibility of the This data packet can trace in reverse the path back to the data
existence of supernodes and the corresponding fraud and consumer. When the intermediate routers receive the data packet,
the low transmission efficiency.
We find out a more suitable way to combine NDN recorded in the PIT. If there is a corresponding entry, they will
network and blockchain. Different from the traditional forward the data packet back to the specified interface and store
implementation, we don mic the architecture it in their CS so they can send back the specific content
of bitcoin systems in IP network but propose a new immediately next time. Each data packet is cryptographically
design. Our design takes advantage of the superiority of signed, enabling the receiver to check the integrity and
content distribution in NDN. By using our naming rules, provenance of the data. There might be several producers to one
nodes fetch data from individual participants and send consumer at the same time. Each producer has the same data
out self-state information at the same time. packet that the consumer request for, so what is noteworthy is
that NDN data delivery naturally supports multicast delivery and
We analyze the advantages of our blockchain system content distribution function [8].
over NDN comparing to those over IP. BlockNDN
resolves the problem of systems over IP that we have B. Blockchain Background
summarized. Our design provides completely On account of its particular subversiveness, blockchain even
decentralized systems and simplifies system architecture. draws more attention than bitcoin-a peer to peer electronic cash
It can decrease the broadcast overhead. Our system also system, from which it comes. The exclusive innovation worked
provides more anonymity to prevent both eavesdropping as a data structure in bitcoin at the very beginning, being like an
and traffic analysis. engine of the credit and security system of the bitcoin. Now it
works as an infrastructure of distributed peers, providing
The rest of our paper is organized as follows. We give a brief services to different industries like banking, healthcare, supply
review of NDN and blockchain , and describe the chain and Internet of Things.
design of our system, BlockNDN, in Section . The Section
introduces how we In brief, the blockchain data structure is a linked list of a
implement our system. In the end, we discuss the future work series of blocks that is built with hash pointers (Fig. 2). Each
and conclude the paper in Section . block contains a list of data units called transactions and a
message digest of the previous block to ensure that the previous
II. OVERVIEW OF NDN AND BLOCKCHAIN transactions will not be tampered [9]. Every blockchain system
A. Named Data Networking Background should resolve consensus problem which means all the
participator should assent to the order of blocks and have the
NDN is a completely new networking model compared with same capacity of knowledge about the blockchain of the system.
TCP/IP. Data is the most important entity in NDN. They are There are many consensus schemes such as PoW, Proof-of-
identified and retrieved with data names following a hierarchical Stake (PoS) [10], PBFT [11] and RPCA [12]. Different schemes
naming mechanism which is similar to the URI naming structure will work in different scenarios. Distributed storage solution
in current IP network. NDN architecture has two basic should also be designed.
communication units: interest packets and data packets, both of
which carry data names. Fig. 1 presents the three key data We introduce a bitcoin blockchain here as a concrete
structure: Forwarding Interest Base (FIB), Content Store (CS) example to show how blockchain works. Bitcoin blockchain

76
The naming convention.

Bitcoin addresses directly to IP data [13] disappears, which can


prevent both eavesdropping and traffic analysis. There are two
Overview.
different naming rules for information exchange in different
system uses PoW as consensus algorithm and nodes maintain the scenarios. Every name consists of three parts: the application
whole blockchain in their memories. Following demonstrates fixed prefix, the function prefix and the digest prefix.
how a bitcoin blockchain system works. The naming rules of BlockNDN are divided into two classes:
Nodes generate transactions and broadcast them to all one for application data names and one for sync data names. We
users. Other nodes called miners collect these name every block in the system blockchain following getblock
transactions into a block and try to make it valid. Miners naming rule. Each name carries the message digest of the
should complete their PoW by changing the nonce values previous block which the (the block) links to. As
of the blocks repeatedly until the hash results satisfy the blockchain structure is linear, a name prefix carrying the
system requirements. A Node creates a new block once hash value of the previous block shows the place of the block in
it satisfies the PoW, and this block cannot be falsified the chain. We follow the same principle in IP that uses digests
without redoing the whole work which has been done in to mark block height in blockchains rather than numerical
the system. symbol. We define the first block in blockchain, the block at
height 0 or depth 0 as well. All users know the hash value of the
When a new valid block is generated, the creator first block which is shown in Fig. 4(a). Part one of this getblock
broadcasts it to the network. Every node checks the PoW data name is the application name so that the interests can be
and accepts this block only if all transactions in it are forwarded towards the producers directly. The purpose of the
valid. After the acceptance, the block gets stored and part two indicates that it is used for naming blocks and that the
another new block created later will chain after it. digest of the previous block is shown in the third part. Interests
with the name in Fig. 4(a) suggest that the consumers have
Eventually, every receiver maintains a chain of the requested for the block at height 1.
blocks recording all the transaction chronologically. The
longest chain in the network is the legitimate bitcoin Similarly, the name for sync data also consists of three parts
blockchain (the best blockchain). (Fig. 4(b)). However, the third part serves the purpose of
carrying the digests of the highest blocks in own
If the whole system behaves correctly, the longest bitcoin memories. This kind of interests following this naming rule can
blockchains of users are same, which indicates that nodes arrive be sent as synchronous requests when the requesters are new
at a consensus. ones or have been disconnected from the network for quite a
III. BLOCKNDN DESIGN while, in which case it becomes time-consuming to ask for
blocks one by one. The responders then extract the current state
A. Overview digests from those interests and return corresponding data
According to different requirements, different blockchain packets when they compare to their own blockchain.
systems are designed. Given the popularity of bitcoin, we decide It is noteworthy that several new blocks can be generated
to implement a bitcoin-like blockchain system named simultaneously, which can cause blockchain forks in the system.
BlockNDN. Our system is totally decentralized. Initialization is This problem can be solved with the exclude filter [14] like what
complete after nodes download all blocks from the first to the is done in the ChronoSync. Exclude filter is a kind of selector
latest one. Meanwhile, miners (Fig. 3) encapsulate data into that can be sent along with the interest to exclude data that the
blocks and broadcast them once they implement the proof-of- requester no longer needs.
work. Listeners will add a new block to the local blockchain if
they have checked the correctness. As a result, the height of the C. Updating Status
best blockchain [2] in the system continues to increase. When consumers are new ones or have been disconnected
from the network for quite a while, it will take a long time to ask
B. Naming Convention
for data by sending getblock interests one by one. We have
Similar to the URI naming structure, data names in NDN described the update of status with sync data exchange before
follow a hierarchical naming mechanism, which is also one of and let us go into more details. A node sends an interest carrying
the most significant aspects of application design. Because NDN an individual state digest and the hash of its highest block to the
have the concept of IP addresses, the ability to map network to inform the latest state of its dataset.A Node providing

77
Updating status between users.

Block structure.

sync service gets the state of this consumer through the getblock
interest name prefix and checks if the name matches with any
block in its own blockchain. If the match is successful, it will

requester gets what it wants in very short time.


Broadcasting a new block.
Dave is a newcomer and Bob has been disconnected for a
long time, both of whom have missed lots of blocks at the current Because of the naming rules, users can get the name of the
state (Fig. 5). Alice and Charles have the whole current blocks s become
the system has generated. Their blockchains are the same. Dave substantially easier. A participant only needs to generate the
and Bob send sync interests to the system with their state at hash of the latest block and sends an interest with name carrying
present. Due to geographic distance, Charles returns blocks from this value. It's also worth pointing out that, different from
height 1 to height 80 to Dave while Alice sends back data packet the Chronosync, users in our system do not construct or maintain
to Bob with blocks from height 41 to 80. digest trees because blockchain is naturally a digest chain and
D. Creating a New Block
In order to simplify the implementation, business IV. DISCUSSION AND ANALYSIS
information will not be carried in transactions like common In this section we discuss the advantages of BlockNDN
blockchain systems. A piece of message in a block just contains compared to the traditional blockchain system in IP network.
mathematical expression so that the result which can be checked
quickly for other users. Messages are also laid out in linear A. Difference Between Logical and Physical Topology
structure from top to bottom, which is different from Merkle In a peer-to-peer system over IP, nodes establish connections
trees in other systems. with other nodes and the volunteer nodes at the application layer,
which form logical topological map. However, messages are
A Miner puts some messages, the hash of the highest block
delivered along the physical paths of physical network which is
in its blockchain and the time into a block according to a
different from those links in the logical map. In our system over
common structure [15] shown in Fig. 6. Then it generates a
NDN, information is propagated along the physical paths
random integer, nonce, and change this value to modify the
directly, which is more efficient than the system with supernodes
digest of the whole block in order to produce a hash below the
in IP network.
threshold level. The threshold in our system is a fix value with
27 leading zeroes in bits, while the threshold in the real bitcoin The broadcasting of blockchain system in IP network is
system should be constantly updated. A Miner can update time complicated. Nodes send messages to every neighbor that they
item and continue its PoW work if all the nonce value fail to get have established the connection with. Their neighbors also
a good result. When users receive blocks, they check validity of follow the same way. In theory, nodes in blockchain are
the PoW and data in blocks sequentially and decide whether to completely homogeneous without any level division, whereas
store the blocks. the implementation is contrived. It is difficult for nodes to find
others participants online without any prompts. To resolve this
E. Broadcasting a New Block
problem, bitcoin in IP network introduces the concept of
Miners should broadcast blocks immediately once they volunteer nodes. The IP addresses of these nodes are
produce PoWs, or else their works and investment will be recommended to users as default connection options to help to
meaningless. We use the broadcast method of the Chronosync discover neighbors and to send neighboring nodes lists. These
protocol [16] for reference. As illustrated in Fig. 7, every node volunteers bring a certain amount of risk to the system because
keeps an outstanding getblock interest to request the next block we cannot ensure that there exists no fraud among them.
automatically when it stores a new block. When all parties have Meanwhile, the logical and the physical topology are different
the same knowledge about the blocks, the system is in a stable between each other, which contributes to poor connectivity and
state. In a stable state, one getblock interest can be transmitted leads to low efficiency of network propagation.
over a link at most in one direction. As soon as some party
generates a new block, the state changes, and the outstanding Fig. 8 is an example. The Black lines and the red lines
interests get satisfied. In Fig. 7, the state change has been caused represent the physical paths and the logical paths between nodes,
data which is multicasted to other two users respectively. Nodes with label of c, d and f are supernodes. Take
following the PIT entries set up in routers by sync interests. node e as an example, it has built a connection with f at the very

78
Among the above, R is the total number of nodes which are
informed of the new message. N is the number of nodes in the
whole network. is the number of forwarding of node i along
logical paths. is the average-physical distance for node i,
which represents the hops node i takes for forwarding the
message to those neighboring nodes according to the logical
topology.
Owing to the consistency of logical topology and physical
topology in NDN network architecture, the average-physical
distance that it takes is 1. A message will be forwarded to all the
other neighboring nodes when a node receives this message for
the first time. Within (2), the broadcast overhead under NDN
The logical topology and the physical topology.
network is as following:

beginning and chooses to establish a link with node a after node


f has sent it a list of IP addresses of other users. When node e has
produced a new block, it will broadcast messages to a and f For every new message M, is the degree of node i in the graph
firstly and inform node j finally when the propagation is done. G of physical links and it represents the number of adjacent
Since NDN has no conception of IP address, nodes do not neighbors of node i in the physical map. N is the number of
need to exchange information over and over again to establish nodes in the whole network. k is the average number of the
relationship with others. They just request data by sending neighbors of G and l is the number of the edges of G. Let us
interests with name prefixes that carry no identities of suppose that all nodes in the system will receive message M
participants. Interest packets are forwarded to producers along eventually.
physical paths. Similar to gossip spreading, the propagation of information
B. Complexity of Data Exchange over IP networks can be divided into two phases [17]. Some
research has shown that the message overhead of the specific
Unlike NDN, the TCP/IP protocol does not provide native
flooding broadcast strategy is lower than gossip spreading
support for multicasting. As a result, extra work should be done
strategy [18], so we suppose the broadcast strategy of the system
for improving system performance.
in IP networks is as simple as possible which sends fewer
To avoid sending block messages to those nodes which have messages than the gossip one. When receiving a new message
already received them from others, data in IP network are not from a neighbor for the first time, a node will send a new
forwarded directly [17]. Instead their existence is announced to message to the other neighbors in logical topology. Since logical
the neighbors by inv messages. The inv message contains a set topology and physical topology are different, the broadcast
of block hashes that have been stored or created by sender and overhead in IP network is as following:
then be forwarded to the neighbors. A node, receiving an inv
message that it has not maintained locally yet, will issue a
getdata message which contains the hashes of the block it needs.
Finally, the transfer of the block will finish via individual block
messages which are sent by the owner. In sum, two nodes will For every new message M, is the degree of node i in the graph
complete information exchange when they complete sending an of logical links and is the average-physical distance of
inv message, a getdata message and a block message. physical topology that node i takes to forward the message to
each neighbor in the logical map.
In NDN, an interest itself indicates the sender does not have
the data at present. The absence of inv messages simplifies the We conclude that the broadcast overhead of the system in
propagation method of the system. NDN or IP network depends on the physical topology and the
logical topology. is only lower than in cases that the
C. Broadcast Overhead connectivity of the physical topology is relatively high. In the
In a blockchain system, peers broadcast their blocks and real bitcoin system over IP, each user has an average of 32 open
transactions to the others. It is important to add that NDN routers connections [17] which is higher than those in physical links. In
with Content Store are equipped with cache capabilities and are fact, the message broadcast over IP is more like gossip spreading
able to increase the network performance. which costs several times more than flooding broadcast strategy.
In general, the broadcast overhead in NDN network is relatively
To evaluate the broadcast overhead in different network lower than in IP network.
architectures, we set up a corresponding scale of evaluation. In
the whole system, the cost of the average node of each V. IMPLEMENTATION
information propagation is defined as the broadcast overhead in We implemented the BlockNDN system using Named Data
this network architecture, which is shown in (1). Networking Forwarding Daemon (NFD) [19], a core component
of the NDN platform for deploying a system on a computer
rather than a simulator, and tested our system on Linux. Every

79
same time, improvements need to be made to resist attacks and
strengthen system users can
develop chat applications, audit systems and others based on our
system. We also believe that blockchain can offer a new
perspective for the progress of NDN in aspects like routing
forwarding, namespacing and key management.
ACKNOWLEDGMENT
This work has been financially supported byShenzhen Key
Fundamental Research Projects (No.JCYJ20160330095313861,
Each node also act as a miner. JCYJ20151030154330711 and JCYJ20151014093505032).

REFERENCES
[1] The IBM blockchain: blockchain overview. [Online]. Available:
http://www.ibm.com/blockchain/what-is-blockchain.html
[2] The bitcoin standard client. [Online]. Available: http://github.com/
bitcoin/bitcoin/
[3] -to-peer
Consulted, 2009.
[4] J.A.D. Donet, C. Prez-Sol, and J. Herrera-Joanco The Bitcoin
P2P Network, The Workshop on Bitcoin Research, vol. 8438, pp. 87-
102, March 2014.
[5] Zhang Relatorio
Time record for the generation of blocks. Tecnico NDN-0001, Xerox Palo Alto Research Center-PARC, 2010.
[6] Van Jacobson, et al., VoCCN:voice-over content-centric networks,
user can run the program in proper working order. In our system, Proceedings of the 2009 ACM Conference on Emerging Networking
each node act as a miner. Nodes kept collecting data and Experiments and Technologies, Rome, Italy, December 2009, 2009.
calculating ceaselessly from the very beginning. They also [7] Zhang Lixia, et al., Named data networking Computer Communication
Review, vol 44, pp. 66-73, July 2014.
requested new blocks at the same time. After a period of time,
we added in node 4 to simulate the join of new members (Fig. [8] Zhenkai Zhu, et al., ACT: audio conference tool over named data
networking, Proceedings of the ACM SIGCOMM Workshop on
9). The node can update their blockchain in short order. Information-Centric Networking (ICN), Toronto, Canada, August 2011.
The difficulty of proof-of-work is configured to a fairly low [9] A. Narayanan, Joseph Bonneau, Edward Felten, Andrew Miller, Steven
Goldfeder, Bitcoin and Cryptocurrency Technologies: A Comprehensive
degree. Miners only need to produce blocks with 27 leading Introduction. Princeton University Press, 2016.
zeros in bits. 27 leading zeros is an empirical value, in which
[10] PPCoin: peer-to-peer crypto-currency with proof-of-stake. [Online].
case nodes can dig blocks in a reasonably-tolerable period of Available: http://archive.org/details/PPCoinPaper
time. Within 597 minutes, 62 blocks were created which formed [11] M. Castro, B. Liskov, Practical Byzantine fault tolerance, Symposium
the best blockchain of the system (Fig. 10). on Operating Systems Design and Implementation USENIX Association,
vol. 20 ,1999 ,pp. 173-176.
VI. CONCLUSION AND FUTURE WORK [12] Executive Summary for Financial Institutions: Ripple Solutions Guide.
In this paper, we present BlockNDN, a bitcoin blockchain [Online]. Available: http://ripple.com/files/ripple_solutions_guide.pdf
decentralized ledger system running on NDN network. Rather [13] Koshy, Philip, D. Koshy, and P. Mcdaniel, An Analysis of Anonymity
in Bi Financial Cryptography and Data
than blindly or mechanically copy the design and completion of Security, Springer Berlin Heidelberg, 2014:469-485.
bitcoin in IP network, we suggested a design method similar to
[14] Ccnx techincal documentation: Ccnx interest message. [Online].
the ChronoSync to achieve decentralization and leverage the Available:http://www.ccnx.org/releases/latest/doc/technical/InterestMes
advantages of NDN architecture. In BlockNDN, the naming sage.html.
rules imply more specific semantics. A sync Interest packet with [15] The Bitcoin Wiki: Block. [Online]. Available:
senders state is forwarded to the producer, which gives a more http://en.bitcoin.it/wiki/Block
simplified method to update dataset. Getblock data names with [16] Zhenkai Zhu, A. Alexander, Let's ChronoSync: Decentralized dataset
digests of the blocks also make retrieval and request more state synchronization in Named Data Networking, Proceedings of the
effortless. The resulting design provides more decentralized IEEE International Conference on Network Protocols(ICNP), Gottingen,
Germany. October 2013, 2013.
systems than those over TCP/IP and makes system architecture
[17] C. Decker, R. Wattenhofer, Information propagation in the Bitcoin
simpler. Proceedings of the IEEE Thirteenth International Conference
Our future work will focus on performance of the system. on Peer-To-Peer Computing (P2P), Trento, Italy, September 2013.
We will port the application code to another NDN platform, [18] Meng Lin, K. Marzullo, a Gossip versus Deterministic
Flooding: Low Message Overhead and High Reliability for Broadcasting
NDNSim, and conduct comparisons between systems over NDN on Small Networks, Proceedings of 14th International Symposium on
and IP. We will also try to answer the question that whether our Distributed Computing (DISC), 1970.
system architecture can reduce the rate of blockchain forks [19] NFD: Named Data Networking Forwarding Daemon. [Online]. Available:
which can lead to substantial divergence within the users. At the http://named-data.net/doc/NFD/current

80

You might also like