Nano Whitepaper

Thursday, May 3, 2018
Download document
Save for later
Add to list

1 Nano: A Feeless Distributed Cryptocurrency Network Colin LeMahieu [email protected] Abstract—Recently, high demand and limited scalability have 1) Poor scalability: Each block in the blockchain can store increased the average transaction times and fees in popular a limited amount of data, which means the system can cryptocurrencies, yielding an unsatisfactory experience. Here we only process so many transactions per second, making introduce Nano, a cryptocurrency with a novel block-lattice ar- chitecture where each account has its own blockchain, delivering spots in a block a commodity. Currently the median near instantaneous transaction speed and unlimited scalability. transaction fee is $10.38 [2]. Each user has their own blockchain, allowing them to update 2) High latency: The average confirmation time is 164 it asynchronously to the rest of the network, resulting in fast minutes [3]. transactions with minimal overhead. Transactions keep track 3) Power inefficient: The Bitcoin network consumes an es- of account balances rather than transaction amounts, allowing aggressive database pruning without compromising security. To timated 27.28TWh per year, using on average 260KWh date, the Nano network has processed 4.2 million transactions per transaction [4]. with an unpruned ledger size of only 1.7GB. Nano’s feeless, Bitcoin, and other cryptocurrencies, function by achieving split-second transactions make it the premier cryptocurrency for consensus on their global ledgers in order to verify legitimate consumer transactions. transactions while resisting malicious actors. Bitcoin achieves Index Terms—cryptocurrency, blockchain, Nano, distributed consensus via an economic measure called Proof of Work ledger, digital, transactions (PoW). In a PoW system participants compete to compute a number, called a nonce, such that the hash of the entire block is in a target range. This valid range is inversely proportional I. I NTRODUCTION to the cumulative computation power of the entire Bitcoin INCE the implementation of Bitcoin in 2009, there has S been a growing shift away from traditional, government- backed currencies and financial systems towards modern pay- network in order to maintain a consistent average time taken to find a valid nonce. The finder of a valid nonce is then allowed to add the block to the blockchain; therefore, those who ments systems based on cryptography, which offer the ability exhaust more computational resources to compute a nonce to store and transfer funds in a trustless and secure manner play a greater role in the state of the blockchain. PoW provides [1]. In order to function effectively, a currency must be resistance against a Sybil attack, where an entity behaves as easily transferable, non-reversible, and have limited or no fees. multiple entities to gain additional power in a decentralized The increased transaction times, large fees, and questionable system, and also greatly reduces race conditions that inherently network scalability have raised questions about the practicality exist while accessing a global data-structure. of Bitcoin as an everyday currency. An alternative consensus protocol, Proof of Stake (PoS), In this paper, we introduce Nano, a low-latency cryp- was first introduced by Peercoin in 2012 [5]. In a PoS system, tocurrency built on an innovative block-lattice data structure participants vote with a weight equivalent to the amount offering unlimited scalability and no transaction fees. Nano of wealth they possess in a given cryptocurrency. With this by design is a simple protocol with the sole purpose of being arrangement, those who have a greater financial investment are a high-performance cryptocurrency. The Nano protocol can given more power and are inherently incentivized to maintain run on low-power hardware, allowing it to be a practical, the honesty of the system or risk losing their investment. PoS decentralized cryptocurrency for everyday use. does away with the wasteful computation power competition, Cryptocurrency statistics reported in this paper are accurate only requiring light-weight software running on low power as of publication date. hardware. The original Nano (RaiBlocks) paper and first beta imple- mentation were published in December, 2014, making it one II. BACKGROUND of the first Directed Acyclic Graph (DAG) based cryptocur- In 2008, an anonymous individual under the pseudonym rencies [6]. Soon after, other DAG cryptocurrencies began Satoshi Nakamoto published a whitepaper outlining the to develop, most notably DagCoin/Byteball and IOTA [7], world’s first decentralized cryptocurrency, Bitcoin [1]. A key [8]. These DAG-based cryptocurrencies broke the blockchain innovation brought about by Bitcoin was the blockchain, a mold, improving system performance and security. Byteball public, immutable and decentralized data-structure which is achieves consensus by relying on a “main-chain” comprised used as a ledger for the currency’s transactions. Unfortunately, of honest, reputable and user-trusted “witnesses”, while IOTA as Bitcoin matured, several issues in the protocol made Bitcoin achieves consensus via the cumulative PoW of stacked trans- prohibitive for many applications: actions. Nano achieves consensus via a balance-weighted vote

2 Receive Repeat Observe Quorum Confirm (a) When no conflict is detected, no further overhead is required. Receive Repeat Observe Conflict Vote Confirm (b) In the event of a conflicting transaction, nodes vote for the valid transaction. Fig. 1. Nano requires no additional overhead for typical transactions. In the event of conflicting transactions, nodes must vote for the transaction to keep on conflicting transactions. This consensus system provides C. Ledger quicker, more deterministic transactions while still maintaining The ledger is the global set of accounts where each account a strong, decentralized system. Nano continues this develop- has its own transaction chain (Figure 2). This is a key design ment and has positioned itself as one of the highest performing component that falls under the category of replacing a run-time cryptocurrencies. agreement with a design-time agreement; everyone agrees via signature checking that only an account owner can modify III. NANO C OMPONENTS their own chain. This converts a seemingly shared data- structure, a distributed ledger, in to a set of non-shared ones. Before describing the overall Nano architecture, we define the individual components that make up the system. D. Node A node is a piece of software running on a computer that A. Account conforms to the Nano protocol and participates in the Nano An account is the public-key portion of a digital signature network. The software manages the ledger and any accounts key-pair. The public-key, also referred to as the address, is the node may control, if any. A node may either store the shared with other network participants while the private-key entire ledger or a pruned history containing only the last few is kept secret. A digitally signed packet of data ensures that block of each account’s blockchain. When setting up a new the contents were approved by the private-key holder. One user node it is recommended to verify the entire history and prune may control many accounts, but only one public address may locally. exist per account. IV. S YSTEM OVERVIEW Unlike blockchains used in many other cryptocurrencies, B. Block/Transaction Nano uses a block-lattice structure. Each account has its own The term “block” and “transaction” are often used in- blockchain (account-chain) equivalent to the account’s trans- terchangeably, where a block contains a single transaction. action/balance history (Figure 2). Each account-chain can only Transaction specifically refers to the action while block refers be updated by the account’s owner; this allows each account- to the digital encoding of the transaction. Transactions are chain to be updated immediately and asynchronously to the signed by the private-key belonging to the account on which rest of the block-lattice, resulting in quick transactions. Nano’s the transaction is performed. protocol is extremely light-weight; each transaction fits within the required minimum UDP packet size for being transmitted over the internet. Hardware requirements for nodes are also Account A Account B Account C minimal, since nodes only have to record and rebroadcast Block NA Block NB Block NC blocks for most transactions (Figure 1). The system is initiated with a genesis account containing Account A Account B Account C the genesis balance. The genesis balance is a fixed quantity Block NA − 1 Block NB − 1 Block NC − 1 and can never be increased. The genesis balance is divided and sent to other accounts via send transactions registered on the .. .. .. genesis account-chain. The sum of the balances of all accounts . . . will never exceed the initial genesis balance which gives the system an upper bound on quantity and no ability to increase Account A Account B Account C it. Block 1 Block 1 Block 1 This section will walk through how different types of transactions are constructed and propagated throughout the network. Account A Account B Account C Block 0 Block 0 Block 0 A. Transactions Fig. 2. Each account has its own blockchain containing the account’s balance Transferring funds from one account to another requires two history. Block 0 must be an open transaction (Section IV-B) transactions: a send deducting the amount from the sender’s

3 .. .. .. . . . transaction, it encodes its accumulated balance and these nodes only need to keep track of the latest block, which allows them to discard historical data while maintaining correctness. Even with a focus on design-time agreements, there is a R S R delay window when validating transactions due to identifying and handling bad actors in the network. Since agreements in Nano are reached quickly, on the order of milliseconds to seconds, we can present the user with two familiar categories R S R of incoming transactions: settled and unsettled. Settled transac- tions are transactions where an account has generated receive blocks. Unsettled transactions have not yet been incorporated in to the receiver’s cumulative balance. This is a replacement S R for the more complex and unfamiliar confirmations metric in Time other cryptocurrencies. B. Creating an Account S S To create an account, you need to issue an open transaction (Figure 4). An open transaction is always the first transaction of every account-chain and can be created upon the first receipt A B C of funds. The account field stores the public-key (address) derived from the private-key that is used for signing. The Fig. 3. Visualization of the block-lattice. Every transfer of funds requires a source field contains the hash of the transaction that sent the send block (S) and a receive block (R), each signed by their account-chain’s owner (A,B,C) funds. On account creation, a representative must be chosen to vote on your behalf; this can be changed later (Section IV-F). The account can declare itself as its own representative. balance and a receive adding the amount to the receiving account’s balance (Figure 3). open { Transferring amounts as separate transactions in the sender’s account: DC04354B1...AE8FA2661B2, and receiver’s accounts serves a few important purposes: source: DC1E2B3F7C...182A0E26B4A, 1) Sequencing incoming transfers that are inherently asyn- representative: xrb_1anr...posrs, chronous. work: 0000000000000000, 2) Keeping transactions small to fit in UDP packets. type: open, 3) Facilitating ledger pruning by minimizing the data foot- signature: 83B0...006433265C7B204 print. } 4) Isolating settled transactions from unsettled ones. More than one account transferring to the same destination Fig. 4. Anatomy of an open transaction account is an asynchronous operation; network latency and the sending accounts not necessarily being in communication with each other means there is no universally agreeable way C. Account Balance to know which transaction happened first. Since addition is The account balance is recorded within the ledger itself. associative, the order the inputs are sequenced does not matter, Rather than recording the amount of a transaction, verification and hence we simply need a global agreement. This is a key (Section IV-I) requires checking the difference between the design component that converts a run-time agreement in to a balance at the send block and the balance of the preceding design-time agreement. The receiving account has control over block. The receiving account may then increment the previous deciding which transfer arrived first and is expressed by the balance as measured into the final balance given in the new signed order of the incoming blocks. receive block. This is done to improve processing speed If an account wants to make a large transfer that was when downloading high volumes of blocks. When requesting received as a set of many small transfers, we want to represent account history, amounts are already given. this in a way that fits within a UDP packet. When a receiving account sequences input transfers, it keeps a running total of its account balance so that at any time it has the ability to D. Sending From an Account transfer any amount with a fixed size transaction. This differs To send from an address, the address must already have an from the input/output transaction model used by Bitcoin and existing open block, and therefore a balance (Figure 5). The other cryptocurrencies. previous field contains the hash of the previous block in the Some nodes are uninterested in expending resources to store account-chain. The destination field contains the account for an account’s full transaction history; they are only interested funds to be sent to. A send block is immutable once confirmed. in each account’s current balance. When an account makes a Once broadcasted to the network, funds are immediately

4 deducted from the balance of the senders account and wait change { as pending until the receiving party signs a block to accept previous: DC04354B1...AE8FA2661B2, these funds. Pending funds should not be considered awaiting representative: xrb_1anrz...posrs, confirmation, as they are as good as spent from the senders work: 0000000000000000, account and the sender cannot revoke the transaction. type: change, signature: 83B0...006433265C7B204 send { } previous: 1967EA355...F2F3E5BF801, balance: 010a8044a0...1d49289d88c, Fig. 7. Anatomy of a change transaction destination: xrb_3w...m37goeuufdp, work: 0000000000000000, type: send, cause a conflicting view on the status of an account and must signature: 83B0...006433265C7B204 be resolved. Only the account’s owner has the ability to sign } blocks into their account-chain, so a fork must be the result of poor programming or malicious intent (double-spend) by the Fig. 5. Anatomy of a send transaction account’s owner. Account A E. Receiving a Transaction Block i + 2 To complete a transaction, the recipient of sent funds must Account A Account A create a receive block on their own account-chain (Figure 6). Block i Block i + 1 The source field references the hash of the associated send Account A transaction. Once this block is created and broadcasted, the Block i + 2 accounts balance is updated and the funds have officially moved into their account. Fig. 8. A fork occurs when two (or more) signed blocks reference the same previous block. Older blocks are on the left; newer blocks are on the right receive { previous: DC04354B1...AE8FA2661B2, Upon detection, a representative will create a vote referenc- source: DC1E2B3F7C6...182A0E26B4A, ing the block b̂i in its ledger and broadcast it to the network. work: 0000000000000000, The weight of a node’s vote, wi , is the sum of the balances of type: receive, all accounts that have named it as its representative. The node signature: 83B0...006433265C7B204 will observe incoming votes from the other M online repre- } sentatives and keep a cumulative tally for 4 voting periods, 1 minute total, and confirm the winning block (Equation 1). Fig. 6. Anatomy of a receive transaction X M v(bj ) = wi ✶b̂i =bj (1) F. Assigning a Representative i=1 Account holders having the ability to choose a representa- b∗ = arg max v(bj ) (2) bj tive to vote on their behalf is a powerful decentralization tool that has no strong analog in Proof of Work or Proof of Stake The most popular block b∗ will have the majority of the protocols. In conventional PoS systems, the account owner’s votes and will be retained in the node’s ledger (Equation 2). node must be running to participate in voting. Continuously The block(s) that lose the vote are discarded. If a representative running a node is impractical for many users; giving a rep- replaces a block in its ledger, it will create a new vote with resentative the power to vote on an account’s behalf relaxes a higher sequence number and broadcast the new vote to the this requirement. Account holders have the ability to reassign network. This is the only scenario where representatives vote. consensus to any account at any time. A change transaction In some circumstances, brief network connectivity issues changes the representative of an account by subtracting the may cause a broadcasted block to not be accepted by all vote weight from the old representative and adding the weight peers. Any subsequent block on this account will be ignored to the new representative (Figure 7). No funds are moved in as invalid by peers that did not see the initial broadcast. A this transaction, and the representative does not have spending rebroadcast of this block will be accepted by the remaining power of the account’s funds. peers and subsequent blocks will be retrieved automatically. Even when a fork or missing block occurs, only the accounts G. Forks and Voting referenced in the transaction are affected; the rest of the A fork occurs when j signed blocks b1 , b2 , . . . , bj claim network proceeds with processing transactions for all other the same block as their predecessor (Figure 8). These blocks accounts.

5 H. Proof of Work B. Transaction Flooding All four transaction types have a work field that must be A malicious entity could send many unnecessary but valid correctly populated. The work field allows the transaction transactions between accounts under its control in an attempt creator to compute a nonce such that the hash of the nonce to saturate the network. With no transaction fees they are concatenated with the previous field in receive/send/change able to continue this attack indefinitely. However, the PoW transactions or the account field in an open transaction is required for each transaction limits the transaction rate the below a certain threshold value. Unlike Bitcoin, the PoW in malicious entity could generate without significantly investing Nano is simply used as an anti-spam tool, similar to Hashcash, in computational resources. Even under such an attack in an and can be computed on the order of seconds [9]. Once a attempt to inflate the ledger, nodes that are not full historical transaction is sent, the PoW for the subsequent block can nodes are able to prune old transactions from their chain; this be precomputed since the previous block field is known; this clamps the storage usage from this type of attack for almost will make transactions appear instantaneous to an end-user so all users. long as the time between transactions is greater than the time required to compute the PoW. C. Sybil Attack An entity could create hundreds of Nano nodes on a single I. Transaction Verification machine; however, since the voting system is weighted based For a block to be considered valid, it must have the on account balance, adding extra nodes in to the network following attributes: will not gain an attacker extra votes. Therefore there is no 1) The block must not already be in the ledger (duplicate advantage to be gained via a Sybil attack. transaction). 2) Must be signed by the account’s owner. D. Penny-Spend Attack 3) The previous block is the head block of the account- chain. If it exists but is not the head, it is a fork. A penny-spend attack is where an attacker spends infinites- 4) The account must have an open block. imal quantities to a large number of accounts in order to 5) The computed hash meets the PoW threshold require- waste the storage resources of nodes. Block publishing is rate- ment. limited by the PoW, so this limits the creation of accounts If it is a receive block, check if the source block hash is and transactions to a certain extent. Nodes that are not full pending, meaning it has not already been redeemed. If it is a historical nodes can prune accounts below a statistical metric send block, the balance must be less than the previous balance. where the account is most likely not a valid account. Finally, Nano is tuned to use minimal permanent storage space, so V. ATTACK V ECTORS space required to store one additional account is proportional to the size of an open block + indexing = 96B + 32B = 128B. Nano, like all decentralized cryptocurrencies, may be at- This equates to 1GB being able to store 8 million penny-spend tacked by malicious parties for attempted financial gain or account. If nodes wanted to prune more aggressively, they can system demise. In this section we outline a few possible attack calculate a distribution based on access frequency and delegate scenarios, the consequences of such an attack, and how Nano’s infrequently used accounts to slower storage. protocol takes preventative measures. E. Precomputed PoW Attack A. Block Gap Synchronization In Section IV-G, we discussed the scenario where a block Since the owner of an account will be the only entity may not be properly broadcasted, causing the network to adding blocks to the account-chain, sequential blocks can be ignore subsequent blocks. If a node observes a block that does computed, along with their PoW, before being broadcasted not have the referenced previous block, it has two options: to the network. Here the attacker generates a myriad of sequential blocks, each of minimal value, over an extended 1) Ignore the block as it might be a malicious garbage period of time. At a certain point, the attacker performs a block. Denial of Service (DoS) by flooding the network with lots of 2) Request a resync with another node. valid transactions, which other nodes will process and echo In the case of a resync, a TCP connection must be formed as quickly as possible. This is an advanced version of the with a bootstrapping node to facilitate the increased amount transaction flooding described in Section V-B. Such an attack of traffic a resync requires. However, if the block was actually would only work briefly, but could be used in conjunction with a bad block, then the resync was unnecessary and needlessly other attacks, such as a >50% Attack (Section V-F) to increase increased traffic on the network. This is a Network Amplifi- effectiveness. Transaction rate-limiting and other techniques cation Attack and results in a denial-of-service. are currently being investigated to mitigate attacks. To avoid unnecessary resyncing, nodes will wait until a certain threshold of votes have been observed for a potentially malicious block before initiating a connection to a bootstrap F. >50% Attack node to synchronize. If a block doesn’t receive enough votes The metric of consensus for Nano is a balance weighted it can be assumed to be junk data. voting system. If an attacker is able to gain over 50% of

6 the voting strength, they can cause the network to oscillate Offline Unsync Attack Active Stake consensus rendering the system broken. An attacker is able to lower the amount of balance they must forfeit by preventing Fig. 9. A potential voting arrangement that could lower 51% attack require- good nodes from voting through a network DoS. Nano takes ments. the following measures to prevent such an attack: 1) The primary defense against this type of attack is voting- If an attacker is able to cause Stake >Active by a combina- weight being tied to investment in the system. An tion of these circumstances, they would be able to successfully account holder is inherently incentivized to maintain flip votes on the ledger at the expense of their stake. We the honesty of the system to protect their investment. can estimate how much this type of attack could cost by Attempting to flip the ledger would be destructive to the examining the market cap of other systems. If we estimate system as a whole which would destroy their investment. 33% of representatives are offline or attacked via DoS, an 2) The cost of this attack is proportional to the market attacker would need to purchase 33% of the market cap in capitalization of Nano. In PoW systems, technology can order to attack the system via voting. be invented that gives disproportionate control compared to monetary investment and if the attack is successful, this technology could be repurposed after the attack is G. Bootstrap Poisoning complete. With Nano the cost of attacking the system The longer an attacker is able to hold an old private-key scales with the system itself and if an attack were to with a balance, the higher the probability that balances that be successful the investment in the attack cannot be existed at that time will not have participating representatives recovered. because their balances or representatives have transferred to 3) In order to maintain the maximum quorum of voters, the newer accounts. This means if a node is bootstrapped to an next line of defense is representative voting. Account old representation of the network where the attacker has a holders who are unable to reliably participate in voting quorum of voting stake compared to representatives at that for connectivity reasons can name a representative who point in time, they would be able to oscillate voting decisions can vote with the weight of their balance. Maximizing to that node. If this new user wanted to interact with anyone the number and diversity of representatives increases besides the attacking node all of their transactions would be network resiliency. denied since they have different head blocks. The net result 4) Forks in Nano are never accidental, so nodes can make is nodes can waste the time of new nodes in the network policy decisions on how to interact with forked blocks. by feeding them bad information. To prevent this, nodes can The only time non-attacker accounts are vulnerable to be paired with an initial database of accounts and known- block forks is if they receive a balance from an attacking good block heads; this is a replacement for downloading the account. Accounts wanting to be secure from block forks database all the way back to the genesis block. The closer can wait a little or a lot longer before receiving from the download is to being current, the higher the probability an account who generated forks or opt to never receive of accurately defending against this attack. In the end, this at all. Receivers could also generate separate accounts attack is probably no worse than feeding junk data to nodes to use when receiving funds from dubious accounts in while bootstrapping, since they wouldn’t be able to transact order to insulate other accounts. with anyone who has a contemporary database. 5) A final line of defense that has not yet been implemented is block cementing. Nano goes to great lengths to settle VI. I MPLEMENTATION block forks quickly via voting. Nodes could be config- ured to cement blocks, which would prevent them from Currently the reference implementation is implemented in being rolled back after a certain period of time. The C++ and has been producing releases since 2014 on Github network is sufficiently secured through focusing on fast [10]. settling time to prevent ambiguous forks. A more sophisticated version of a > 50% attack is detailed A. Design Features in Figure 9. “Offline” is the percentage of representatives who The Nano implementation adheres to the architecture stan- have been named but are not online to vote. “Stake” is the dard outlined in this paper. Additional specifications are de- amount of investment the attacker is voting with. “Active” scribed here. is representatives that are online and voting according to the 1) Signing Algorithm: Nano uses a modified ED25519 protocol. An attacker can offset the amount of stake they must elliptic curve algorithm with Blake2b hashing for all digital forfeit by knocking other voters offline via a network DoS signatures [11]. ED25519 was chosen for fast signing, fast attack. If this attack can be sustained, the representatives being verification, and high security. attacked will become unsynchronized and this is demonstrated 2) Hashing Algorithm: Since the hashing algorithm is only by “Unsync.” Finally, an attacker can gain a short burst in used to prevent network spam, the algorithm choice is less relative voting strength by switching their Denial of Service important when compared to mining-based cryptocurrencies. attack to a new set of representatives while the old set is re- Our implementation uses Blake2b as a digest algorithm against synchronizing their ledger, this is demonstrated by “Attack.” block contents [12].

7 3) Key Derivation Function: In the reference wallet, keys 4) Light: A light node is also a trusting node that only are encrypted by a password and the password is fed through observes traffic for accounts in which it is interested allowing a key derivation function to protect against ASIC cracking minimal network usage. attempts. Presently Argon2 [13] is the winner of the only 5) Bootstrap: A bootstrap node serves up parts or all of the public competition aimed at creating a resilient key derivation ledger for nodes that are bringing themselves online. This is function. done over a TCP connection rather than UDP since it involves 4) Block Interval: Since each account has its own a large amount of data that requires advanced flow control. blockchain, updates can be performed asynchronous to the state of network. Therefore there are no block intervals and B. Disk Capacity transactions can be published instantly. 5) UDP Message Protocol: Our system is designed to Depending on the user demands, different node configura- operate indefinitely using the minimum amount of computing tions require different storage requirements. resources as possible. All messages in the system were de- 1) Historical: A node interested in keeping a full historical signed to be stateless and fit within a single UDP packet. This record of all transactions will require the maximum amount also makes it easier for lite peers with intermittent connectivity of storage. to participate in the network without reestablishing short-term 2) Current: Due to the design of keeping accumulated TCP connections. TCP is used only for new peers when they balances with blocks, nodes only need to keep the latest want to bootstrap the block chains in a bulk fashion. or head blocks for each account in order to participate in Nodes can be sure their transaction was received by the consensus. If a node is uninterested in keeping a full history network by observing transaction broadcast traffic from other it can opt to keep only the head blocks. nodes as it should see several copies echoed back to itself. 3) Light: A light node keeps no local ledger data and only participates in the network to observe activity on accounts in which it is interested or optionally create new transactions with B. IPv6 and Multicast private keys it holds. Building on top of connection-less UDP allows future implementations to use IPv6 multicast as a replacement for C. CPU traditional transaction flooding and vote broadcast. This will reduce network bandwidth consumption and give more policy 1) Transaction Generating: A node interested in creating flexibility to nodes going forward. new transactions must produce a Proof of Work nonce in order to pass Nano’s throttling mechanism. Computation of various hardware is benchmarked in Appendix A. C. Performance 2) Representative: A representative must verify signatures At the time of this writing, 4.2 million transactions have for blocks, votes, and also produce its own signatures to been processed by the Nano network, yielding a blockchain participate in consensus. The amount of CPU resources for a size of 1.7GB. Transaction times are measured on the order representative node is significantly less than transaction gener- of seconds. A current reference implementation operating on ating and should work with any single CPU in a contemporary commodity SSDs can process over 10,000 transactions per computer. second being primarily IO bound. 3) Observer: An observer node doesn’t generate its own votes. Since signature generation overhead is minimal, the VII. R ESOURCE U SAGE CPU requirements are almost identical to running a represen- This is an overview of resources used by a Nano node. tative node. Additionally, we go over ideas for reducing resource usage for specific use cases. Reduced nodes are typically called light, VIII. C ONCLUSION pruned, or simplified payment verification (SPV) nodes. In this paper we presented the framework for a trustless, feeless, low-latency cryptocurrency that utilizes a novel block- A. Network lattice structure and delegated Proof of Stake voting. The The network activity of a node is dependent on how much network requires minimal resources, no high-power mining the node contributes towards the health of a network. hardware, and can process high transaction throughput. All 1) Representative: A representative node requires maxi- of this is achieved by having individual blockchains for each mum network resources as it observes vote traffic from other account, eliminating access issues and inefficiencies of a representatives and publishes its own votes. global data-structure. We identified possible attack vectors on 2) Trustless: A trustless node is similar to a representative the system and presented arguments on how Nano is resistant node but is only an observer, it doesn’t contain a representative to these forms of attacks. account private key and does not publish votes of its own. 3) Trusting: A trusting node observes vote traffic from A PPENDIX A one representative it trusts to correctly perform consensus. P OW H ARDWARE BENCHMARKS This cuts down on the amount of inbound vote traffic from As mentioned previously, the PoW in Nano is to reduce representatives going to this node. network spam. Our node implementation provides acceleration

8 that can take advantage of OpenCL compatible GPUs. Table I provides a real-life benchmark comparison of various hard- ware. Currently the PoW threshold is fixed, but an adaptive threshold may be implemented as average computing power progresses. TABLE I H ARDWARE P OW P ERFORMANCE Device Transactions Per Second Nvidia Tesla V100 (AWS) 6.4 Nvidia Tesla P100 (Google,Cloud) 4.9 Nvidia Tesla K80 (Google,Cloud) 1.64 AMD RX 470 OC 1.59 Nvidia GTX 1060 3GB 1.25 Intel Core i7 4790K AVX2 0.33 Intel Core i7 4790K,WebAssembly (Firefox) 0.14 Google Cloud 4 vCores 0.14-0.16 ARM64 server 4 cores (Scaleway) 0.05-0.07 ACKNOWLEDGMENT We would like to thank Brian Pugh for compiling and formatting this paper. R EFERENCES [1] S. Nakamoto, “Bitcoin: A peer-to-peer electronic cash system,” 2008. [Online]. Available: http://bitcoin.org/bitcoin.pdf [2] “Bitcoin median transaction fee historical chart.” [Online]. Avail- able: https://bitinfocharts.com/comparison/bitcoin-median transaction fee.html [3] “Bitcoin average confirmation time.” [Online]. Available: https: //blockchain.info/charts/avg-confirmation-time [4] “Bitcoin energy consumption index.” [Online]. Available: https: //digiconomist.net/bitcoin-energy-consumption [5] S. King and S. Nadal, “Ppcoin: Peer-to-peer crypto-currency with proof-of-stake,” 2012. [Online]. Available: https://peercoin.net/assets/ paper/peercoin-paper.pdf [6] C. LeMahieu, “Raiblocks distributed ledger network,” 2014. [7] Y. Ribero and D. Raissar, “Dagcoin whitepaper,” 2015. [8] S. Popov, “The tangle,” 2016. [9] A. Back, “Hashcash - a denial of service counter-measure,” 2002. [Online]. Available: http://www.hashcash.org/papers/hashcash.pdf [10] C. LeMahieu, “Raiblocks,” 2014. [Online]. Available: https://github. com/clemahieu/raiblocks [11] D. J. Bernstein, N. Duif, T. Lange, P. Shwabe, and B.-Y. Yang, “High-speed high-security signatures,” 2011. [Online]. Available: http://ed25519.cr.yp.to/ed25519-20110926.pdf [12] J.-P. Aumasson, S. Neves, Z. Wilcox-O’Hearn, and C. Winnerlein, “Blake2: Simpler, smaller, fast as md5,” 2012. [Online]. Available: https://blake2.net/blake2.pdf [13] A. Biryukov, D. Dinu, and D. Khovratovich, “Argon2: The memory- hard function for password hashing and other applications,” 2015. [Online]. Available: https://password-hashing.net/argon2-specs.pdf