bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 3쪽 중 2쪽
es in the seed anchor
set, then the accepted state transitions must have been marked as accepted by at least one correct node.
This state discovery process is also used for network discovery. The membership set of the network is
defined on the validator chain. Therefore, synchronizing with the validator chain allows the node to discover205
the current set of validators. The validator chain will be discussed further in the next section.
3.3 Sybil Control and Membership
Consensus protocols provide their security guarantees under the assumption that up to a threshold number
of members in the system could be adversarial. A Sybil attack, wherein a node cheaply floods the network
with malicious identities, can trivially invalidate these guarantees. Fundamentally, such an attack can only be210
deterred by trading off presence with proof of a hard-to-forge resource [3]. Past systems have explored the use
of Sybil deterrence mechanisms that span proof-of-work (PoW), proof-of-stake (PoS), proof-of-elapsed-time
(POET), proof-of-space-and-time (PoST), and proof-of-authority (PoA).
At their core, all of these mechanisms serve an identical function: they require that each participant have
some “skin in the game” in the form of some economic commitment, which in turn provides an economic215
barrier against misbehavior by that participant. All of them involve a form of stake, whether it is in the form
of mining rigs and hash power (PoW), disk space (PoST), trusted hardware (POET), or an approved identity
(PoA). This stake forms the basis of an economic cost that participants must bear to acquire a voice. For
instance, in Bitcoin, the ability to contribute valid blocks is directly proportional to the hash-power of the
proposing participant. Unfortunately, there has also been substantial confusion between consensus protocols220
8 Kevin Sekniqi, Daniel Laine, Stephen Buttolph, and Emin G¨ un Sirer
versus Sybil control mechanisms. We note that the choice of consensus protocols is, for the most part,
orthogonal to the choice of the Sybil control mechanism. This is not to say that Sybil control mechanisms are
drop-in-replacements for each other, since a particular choice might have implications about the underlying
guarantees of the consensus protocol. However, the Snow* family can be coupled with many of these known
mechanisms, without significant modification.225
Ultimately, for security and to ensure that the incentives of participants are aligned for the benefit of
the network, $AVAX choose PoS to the core Sybil control mechanism. Some forms of stake are inherently
centralized: mining rig manufacturing (PoW), for instance, is inherently centralized in the hands of a few
people with the appropriate know-how and access to the dozens of patents required for competitive VLSI
manufacturing. Furthermore, PoW mining leaks value due to the large yearly miner subsidies. Similarly,230
disk space is most abundantly owned by large datacenter operators.Further, all sybil control mechanisms
that accrue ongoing costs, e.g. electricity costs for hashing, leak value out of the ecosystem, not to mention
destroy the environment. This, in turn, reduces the feasibility envelope for the token, wherein an adverse
price move over a small time frame can render the system inoperable. Proof-of-work inherently selects for
miners who have the connections to procure cheap electricity, which has little to do with the miners’ ability235
to serialize transactions or their contributions to the overall ecosystem. Among these options, we choose
proof-of-stake, because it is green, accessible, and open to all. We note, however, that while the $AVAX uses
PoS, the Avalanche network enables subnets to be launched with PoW and PoS.
Staking is a natural mechanism for participation in an open network because it enables a direct economic
argument: the probability of success of an attack is directly proportional to a well-defined monetary cost240
function. In other words, the nodes that stake are economically motivated to not engage in behavior that
might hurt the value of their stake. Additionally, this stake does not incur any additional upkeep costs (other
then the opportunity cost of investing in another asset), and has the property that, unlike mining equipment,
is fully consumed if used in a catastrophic attack. For PoW operations, mining equipment can be simply
reused or – if the owner decides to – entirely sold back to the market.245
A node wishing to enter the network can freely do so by first putting up a stake that is immobilized
during the duration of participation in the network. The user determines the amount duration of the stake.
Once accepted, a stake cannot be reverted. The main goal is to ensure that nodes substantially share the
same mostly stable view of the network. We anticipate setting the minimum staking time on the order of a
week.250
Unlike other systems that also propose a PoS mechanism, $AVAX does not make usage of slashing, and
therefore all stake is returned when the staking period expires. This prevents unwanted scenarios such as
a client software or hardware failure leading to a loss of coins. This dovetails with our design philosophy
of building predictable technology: the staked tokens are not at risk, even in the presence of software or
hardware flaws.255
In Avalanche, a node that wants to participate issues a special stake transaction to the validator chain.
Staking transactions name an amount to stake, the staking key of the participant that is staking, the duration,
and the time that validation will start. Once the transaction is accepted, the funds will be locked until the
end of the staking period. The minimal allowed amount is decided and enforced by the system. The stake
amount placed by a participant has implications for both the amount of influence the participant has in the260
Avalanche Platform 2020/06/30 9
consensus process, as well as the reward, as discussed later. The specified staking duration, must be between
δmin and δmax, the minimum and maximum timeframes for which any stake can be locked. As with the
staking amount, the staking period also has implications for the reward in the system. Loss or theft of the
staking key cannot lead to asset loss, as the staking key is used only in the consensus process, not for asset
transfer.265
3.4 Smart Contracts in $AVAX
At launch Avalanche supports standard Solidity-based smart contracts through the Ethereum virtual ma-
chine (EVM). We envision that the platform will support a richer and more powerful set of smart contract
tools, including:
– Smart contracts with off-chain execution and on-chain verification.270
– Smart contracts with parallel execution. Any smart contracts that do not operate on the same state in
any subnet in Avalanche will be able to execute in parallel.
– An improved Solidity, called Solidity++. This new language will support versioning, safe mathematics
and fixed point arithmetic, an improved type system, compilation to LLVM, and just-in-time execution.
If a developer requires EVM support but wants to deploy smart contracts in a private subnet, they275
can spin-up a new subnet directly. This is how Avalanche enables functionality-specific sharding through
the subnets. Furthermore, if a developer requires interactions with the currently deployed Ethereum smart
contracts, they can interact with the Athereum subnet, which is a spoon of Ethereum. Finally, if a developer
requires a different execution environment from the Ethereum virtual machine, they may choose to deploy
their smart contract through a subnet that implements a different execution environment, such as DAML280
or WASM. Subnets can support additional features beyond VM behavior. For example, subnets can enforce
performance requirements for bigger validator nodes that hold smart contracts for longer periods of time, or
validators that hold contract state privately.
4 Governance and The $AVAX Token
4.1 The $AVAX Native T oken285
Monetary Policy The native token, $AVAX, is capped-supply, where the cap is set at 720 , 000, 000 tokens,
with 360, 000, 000 tokens available on mainnet launch. However, unlike other capped-supply tokens which
bake the rate of minting perpetually, $AVAX is designed to react to changing economic conditions. In par-
ticular, the objective of $AVAX’s monetary policy is to balance the incentives of users to stake the token
versus using it to interact with the variety of services available on the platform. Participants in the platform290
collectively act as a decentralized reserve bank. The levers available on Avalanche are staking rewards, fees,
and airdrops, all of which are influenced by governable parameters. Staking rewards are set by on-chain gov-
ernance, and are ruled by a function designed to never surpass the capped supply. Staking can be induced
by increasing fees or increasing staking rewards. On the other hand, we can induce increased engagement
with the Avalanche platform services by lowering fees, and decreasing the staking reward.295
10 Kevin Sekniqi, Daniel Laine, Stephen Buttolph, and Emin G¨ un Sirer
Uses
Payments True decentralized peer-to-peer payments are largely an unrealized dream for the industry due to
the current lack of performance from incumbents. $AVAX is as powerful and easy to use as payments using
Visa, allowing thousands of transactions globally every second, in a fully trustless, decentralized manner.
Furthermore, for merchants worldwide, $AVAX provides a direct value proposition over Visa, namely lower300
fees.
Staking: Securing the System On the Avalanche platform, sybil control is achieved via staking. In order
to validate, a participant must lock up coins, or stake. Validators, sometimes referred to as stakers, are
compensated for their validation services based on staking amount and staking duration, amongst other
properties. The chosen compensation function should minimize variance, ensuring that large stakers do not305
disproportionately receive more compensation. Participants are also not subject to any “luck” factors, as in
PoW mining. Such a reward scheme also discourages the formation of mining or staking pools enabling truly
decentralized, trustless participation in the network.
Atomic swaps Besides providing the core security of the system, the $AVAX token serves as the universal unit
of exchange. From there, the Avalanche platform will be able to support trustless atomic swaps natively on310
the platform enabling native, truly decentralized exchanges of any type of asset directly on Avalanche.
4.2 Governance
Governance is critical to the development and adoption of any platform because – as with all other types
of systems – Avalanche will also face natural evolution and updates. $AVAX provides on-chain governance
for critical parameters of the network where participants are able to vote on changes to the network and315
settle network upgrade decisions democratically. This includes factors such as the minimum staking amount,
minting rate, as well as other economic parameters. This enables the platform to effectively perform dy-
namic parameter optimization through a crowd oracle. However, unlike some other governance platforms
out there, Avalanche does not allow unlimited changes to arbitrary aspects of the system. Instead, only a
pre-determined number of parameters can be modified via governance, rendering the system more predictable320
and increasing safety. Further, all governable parameters are subject to limits within specific time bounds,
introducing hysteresis, and ensuring that the system remains predictable over short time ranges.
A workable process for finding globally acceptable values for system parameters is critical for decentral-
ized systems without custodians. Avalanche can use its consensus mechanism to build a system that allows
anyone to propose special transactions that are, in essence, system-wide polls. Any participating node may325
issue such proposals.
Nominal reward rate is an important parameter that affects any currency, whether digital or fiat. Unfor-
tunately, cryptocurrencies that fix this parameter might face various issues, including deflation or inflation.
To that end, the nominal reward rate is subject to governance, within pre-established boundaries. This will
allow token holders to choose on whether $AVAX is eventually capped, uncapped, or even deflationary.330
Avalanche Platform 2020/06/30 11
Transaction fees, denoted by the setF, are also subject to governance.F is effectively a tuple which de-
scribes the fees associated with the various instructions and transactions. Finally, staking times and amounts
are also governable. The list of these parameters is defined in Figure 1.
– ∆ : Staking amount, denominated in $AVAX. This value defines the minimal stake required to be placed as
bond before participating in the system.
– δmin : The minimal amount of time required for a node to stake into the system.
– δmax : The maximal amount of time a node can stake.
– ρ : (π∆,τδ min)→ R : Reward rate function, also referred to as minting rate, determines the reward a
participant can claim as a function of their staking amount given some number ofπ publicly disclosed nodes
under its ownership, over a period of τ consecutiveδmin timeframes, such that τδ min≤δmax.
– F : the fee structure, which is a set of governable fees parameters that specify costs to various transactions.
Fig. 1. Key non-consensus parameters used in Avalanche. All notation is redefined upon first use.
In line with the principle of predictability in a financial system, governance in $AVAX has hysteresis,
meaning that changes to parameters are highly dependent on their recent changes. There are two limits335
associated with each governable parameter: time and range. Once a parameter is changed using a governance
transaction, it becomes very difficult to change it again immediately and by a large amount. These difficulty
and value constraints relax as more time passes since the last change. Overall, this keeps the system from
changing drastically over a short period of time, allowing users to safely predict system parameters in the
short term, while having strong control and flexibility for the long term.340
5 Discussion
5.1 Optimizations
Pruning Many blockchain platforms, especially those implementing Nakamoto consensus such as Bitcoin,
suffer from perpetual state growth. This is because – by protocol – they have to store the entire history of
transactions. However, in order for a blockchain to grow sustainably, it must be able to prune old history.345
This is especially important for blockchains that support high performance, such as Avalanche.
Pruning is simple in the Snow* family. Unlike in Bitcoin (and similar protocols), where pruning is not
possible per the algorithmic requirements, in $AVAX nodes do not need to maintain parts of the DAG that
are deep and highly committed. These nodes do not need to prove any past history to new bootstrapping
nodes, and therefore simply have to store active state, i.e. the current balances, as well as uncommitted350
transactions.
Client Types Avalanche can support three different types of clients: archival, full, and light. Archival
nodes store the entire history of the $AVAX subnet, the staking subnet, and the smart contract subnet, all the
12 Kevin Sekniqi, Daniel Laine, Stephen Buttolph, and Emin G¨ un Sirer
way to genesis, meaning that these nodes serve as bootstrapping nodes for new incoming nodes. Additionally
these nodes may store the full history of other subnets for which they choose to be validators. Archival355
nodes are typically machines with high storage capabilities that are paid by other nodes when downloading
old state. Full nodes, on the other hand, participate in validation, but instead of storing all history, they
simply store the active state (e.g. current UTXO set). Finally, for those that simply need to interact securely
with the network using the most minimal amount of resources, Avalanche supports light clients which can
prove that some transaction has been committed without needing to download or synchronize history. Light360
clients engage in the repeated sampling phase of the protocol to ensure safe commitment and network wide
consensus. Therefore, light clients in Avalanche provide the same security guarantees as full nodes.
Sharding Sharding is the process of partitioning various system resources in order to increase performance
and reduce load. There are various types of sharding mechanisms. In network sharding, the set of participants
is divided into separate subnetworks as to reduce algorithmic load; in state sharding, participants agree on365
storing and maintaining only specific subparts of the entire global state; lastly, in transaction sharding,
participants agree to separate the processing of incoming transactions.
In Avalanche Borealis , the first form of sharding exists through the subnetworks functionality. For
example, one may launch a gold subnet and another real-estate subnet. These two subnets can exist entirely in
parallel. The subnets interact only when a user wishes to buy real-estate contracts using their gold holdings,370
at which point Avalanche will enable an atomic swap between the two subnets.
5.2 Concerns
Post Quantum Cryptography Post-quantum cryptography has recently gained widespread attention
due to the advances in the development of quantum computers and algorithms. The concern with quantum
computers is that they can break some of the currently deployed cryptographic protocols, specifically digital375
signatures. The Avalanche network model enables any number of VMs, so it supports a quantum-resistant
virtual machine with a suitable digital signature mechanism. We anticipate several types of digital signature
schemes to be deployed, including quantum resistant RLWE-based signatures. The consensus mechanism
does not assume any kind of heavy crypto for its core operation. Given this design, it is straightforward to
extend the system with a new virtual machine that provides quantum secure cryptographic primitives.380
Realistic Adversaries The Avalanche paper [6] provides very strong guarantees in the presence of a
powerful and hostile adversary, known as a round-adaptive adversary in the full point-to-point model. In
other terms, the adversary has full access to the state of every single correct node at all times , knows the
random choices of all correct nodes, as well as can update its own state at any time, before and after the
correct node has the chance to update its own state. Effectively, this adversary is all powerful, except for385
the ability to directly update the state of a correct node or modify the communication between correct
nodes. Nonetheless, in reality, such an adversary is purely theoretical since practical implementations of the
strongest possible adversary are limited at statistical approximations of the network state. Therefore, in
practice, we expect worst-case-scenario attacks to be difficult to deploy.
Avalanche Platform 2020/06/30 13
Inclusion and Equality A common problem in permissionless currencies is that of the “rich getting390
richer”. This is a valid concern, since a PoS system that is improperly implemented may in fact allow
wealth generation to be disproportionately attributed to the already large holders of stake in the system. A
simple example is that of leader-based consensus protocols, wherein a subcommittee or a designated leader
collects all the rewards during its operation, and where the probability of being chosen to collect rewards is
proportional to the stake, accruing strong reward compounding effects. Further, in systems such as Bitcoin,395
there is a “big get bigger” phenomenon where the big miners enjoy a premium over smaller ones in terms
of fewer orphans and less lost work. In contrast, Avalanche employs an egalitarian distribution of minting:
every single participant in the staking protocol is rewarded equitably and proportionally based on stake프로젝트로 돌아가기