bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 19쪽 중 3쪽
new users enter
18
the system, increasing potential future earnings from running Chainlink nodes, the
marginal cost of economic security drops for current and future users. In a regime of
elastic demand, this diminished cost incentivizes additional users to make use of the
network, continuously perpetuating adoption in an ongoing virtuous cycle.
Note: While this whitepaper outlines important elements of our vision for the evolu-
tion of Chainlink, it is informal and includes few detailed technical specifics. We plan to
release focused technical papers on additional features and approaches as they evolve.
Furthermore, it is important to emphasize that many elements of the vision presented
here (scaling improvements, confidentiality technologies, FSS, etc.) can and will be
deployed in preliminary form even before advanced DONs become a basic feature of
Chainlink.
1.3 Organization of this Paper
We present our security model and notation in Section 2 and outline the Decentralized
Oracle Network API in Section 3. In Section 4, we present a number of examples of
applications for which DONs provide an appealing deployment platform. Readers can
learn most of the key concepts of the paper by reading up to this point.
The remainder of the paper contains further details. We describe Fair Sequencing
Services (FSS) in Section 5 and the Transaction-Execution Framework (TEF) in Sec-
tion 6. We describe our approach to trust minimization in Section 7. We consider some
important DON deployment requirements, namely incremental rollout of features, dy-
namic ledger membership, and accountability in Section 8. Finally, in Section 9, we give
an overview of our developing approach to incentive design. We conclude in Section 10.
To help readers who have limited familiarity with the concepts in this paper, we
provide a glossary in Appendix A. We present further detail on the DON interface
and functionality in Appendix B and present some example adapters in Appendix C.
In Appendix D, we describe a cryptographic primitive for trust-minimized data-source
authentication called functional signatures and introduce a new variant called dis-
cretized functional signatures . We discuss some considerations bearing on committee
selection for DONs in Appendix F.
2 Security Model and Goals
A Decentralized Oracle Network is a distinct distributed system that we expect will
initially be implemented typically—although not necessarily—by a committee-based
consensus protocol and run by a set of oracle nodes. A DON is designed primarily
to augment the capabilities of a smart contract on a main chain with oracle reports
and other services, but it can provide those same supporting services to other non-
blockchain systems, and thus need not be associated with a particular main chain.
19
The model and properties we consider are therefore largely independent of the use of
the particular applications of a DON.
2.1 Current Architectural Model
It is important to emphasize that Chainlink today is not a monolithic service, but rather
a permissionless framework within which it is possible to launch distinct, independent
networks of oracle nodes [77]. Networks have heterogeneous sets of node operators and
designs. They may also differ in terms of the types of services they provide, which can
include, e.g., data feeds, Proof of Reserves, verifiable randomness, and so forth. Other
differences can include the degree of decentralization, size of the network in terms of
locked value it supports, and various service-level parameters, such as data frequency
and accuracy.
Chainlink’s permissionless model encourages the growth of an ecosystem in which
providers specialize in the services they are best able to furnish to the community. This
model is likely to result in lower costs to users and higher service quality than a model
that requires all nodes and networks to provide a full range of services, an approach
that can easily devolve into system-wide adoption of the services representing the least
common denominator of resources available to nodes.
As Chainlink evolves toward DON-based designs in Chainlink 2.0, we continue to
support the model of a permissionless, open framework, keeping in view the goal of
providing users with a range of service choices that globally result in the best match
with particular application requirements.
2.2 Consensus Assumptions
We use the term Decentralized Oracle Network to encompass the full functionality of
the oracle system we describe: both the data structure that oracle nodes maintain and
the core API layered on top of it.
We use the term ledger (lower case), denoted by L, to mean the underlying data
structure maintained by a DON and used to support the particular services it provides.
We emphasize that our DON framework does not treatL as a freestanding system like
a blockchain: Its purpose is to support blockchains and other systems. Blockchains are,
of course, one way of realizing a trustworthy ledger, but there are others. We expect
DONs in many cases to realize their underlying ledgers using Byzantine Fault Tolerant
(BFT) systems, which considerably predate blockchains such as Bitcoin [174]. We use
BFT-type notation and properties throughout the paper for convenience, although we
emphasize that DONs can be realized using permissionless consensus protocols.
Conceptually, a ledger L is a bulletin board on which data is linearly ordered.
We view a ledger generally as having a few key properties commonly ascribed to
blockchains [115]. A ledger is:
• Append-only: Data, once added, cannot be removed or modified.
20
• Public: Anyone can read its contents, which are consistent across time in the
view of all users. 4
• Available: The ledger can always be written to by authorized writers and read
by anyone in a timely way.
Alternative properties are possible in the ledger for a DON when realized by a
committee. For instance, ledger write access might be restricted to certain users, as
might read access for some applications, i.e., the ledger need not be public as defined
above. Similarly, ledger rules might permit modification or redaction of data. We don’t
explicitly consider such variants in this paper, however.
The modular design of DONs can support any of a wide variety of modern BFT
protocols, e.g., Hotstuff [231]. The exact choice will depend on trust assumptions and
network characteristics among the oracle nodes. A DON could in principle alternatively
use a highly performant permissionless blockchain for its ledger in its role supporting an
equally scalable layer-2 or blockchain system. Similarly, hybridization is also possible:
The DON could in principle be composed of nodes that are validators in an existing
blockchain, e.g., in Proof-of-Stake systems in which committees are selected to execute
transactions, e.g., [8, 81, 120, 146, 204]. This particular mode of operation requires that
nodes operate in a dual-use manner, i.e., operate both as blockchain nodes and DON
nodes. (See Section 8.2 for a discussion of techniques to ensure continuity in changing
committees and Appendix F for some caveats on random committee selection.)
In practice, in modern BFT algorithms, nodes digitally sign messages on the ledger.
We assume for convenience thatL has an associated public keypkL and that its contents
are signed by the corresponding private key. This general notation applies even when
data onL are signed using threshold signatures.5 Threshold signatures are convenient,
as they enable a persistent identity for a DON even with changes of membership in
the nodes running it. (See Appendix B.1.3.) We thus assume that skL is secret-shared
in a ( k,n )-threshold manner for some security parameter k, e.g., k = 2f + 1 and
n = 3f + 1, where f is the number of potentially faulty nodes. (By choosing k in this
way, we ensure that faulty nodes can neither learn skL nor mount a denial-of-service
attack preventing its use.)
A message on L takes the form M = (m,z ), where m is a string and z a unique
sequential index number. Where applicable, we write messages in the form m =
⟨MessageType : payload⟩. The message type MessageType is syntactic sugar that in-
dicates the function of a particular message.
4In cases where a blockchain without finality realizes a ledger, inconsistency is typically abstracted
away by disregarding insufficiently deep blocks or “pruning” [115].
5In practice, some code bases, e.g., LibraBFT [205], a variant of Hotstuff, have currently adopted
multi-signatures, rather than threshold signatures, trading off reduced communication complexity for
simpler engineering. With some added cost, oracle nodes can append threshold signatures to messages
written toL even if the consensus protocol used for L doesn’t employ them.
21
2.3 Notation
We denote the set of n oracle nodes running the ledger by O = {Oi}ni=1. Such a
set of nodes is often called a committee. For simplicity, we assume that the set of
oracles implementing DON functionality, i.e., services on top of L, is identical with
that maintaining L, but they can be distinct. We let pki denote the public key of
playerOi, and ski the corresponding private key.
Most BFT algorithms require at least n = 3f + 1 nodes, where f is the number of
potentially faulty nodes; remaining nodes are honest, in the sense that they follow the
protocol exactly as specified. We refer to the committee O as honest if it meets this
requirement, i.e., has greater than a 2 /3-fraction of honest nodes. Unless otherwise
stated, we assume that O is honest (and a static model of corruption). We use pkO /
skO interchangeably with pkL / skL, depending on the context.
We letσ = Sigpk[m] denote a signature on message m with respect to pk, i.e., using
corresponding private key sk. Let verify(pk,σ,m )→{ false, true} denote a corre-
sponding signature verification algorithm. (We leave key generation implicit through-
out the paper.)
We use the notation S to denote a data source and S to denote the full set of
nS sources in a given context. We denote by MAINCHAIN a smart-contract enabled
blockchain supported by a DON. We use the term relying contract to denote any smart
contract on MAINCHAIN that communicates with a DON, and use the notation SC to
denote such a contract.
We generally assume that a DON supports a single main chain MAINCHAIN, al-
though it can support multiple such chains, as we show in examples in Section 4. A
DON can and typically will support multiple relying contracts on MAINCHAIN. (As
noted above, a DON can alternatively support non-blockchain services.)
2.4 Note on Trust Models
As noted above, DONs may be built atop committee-based consensus protocols, and we
expect they will commonly use such protocols. There are many strong arguments that
one of the two alternatives, committee-based or permissionless blockchains, provides
stronger security than the other.
It is important to recognize that the security of committee-based vs. permissionless
decentralized systems is incommensurable. Compromising a PoW or a PoS blockchain
via 51% attack requires that an adversary obtain majority resources ephemerally and
potentially anonymously , for example by renting hash power in a PoW system. Such
attacks in practice have already impacted several blockchains [200, 34]. In contrast,
compromising a committee-based system means corrupting a threshold number (typi-
cally one-third) of its nodes, where the nodes may be publicly known, well resourced,
and trustworthy entities.
On the other hand, committee-based systems (as well as “hybrid” permissionless
systems that support committees) can support more functionality than strictly per-
22
missionless systems. This includes the ability to maintain persistent secrets, such as
signing and/or encryption keys—one possibility in our designs.
We emphasize that DONs can in principle be built atop either a committee-based or
permissionless consensus protocol and DON deployers may ultimately choose to adopt
either approach.
Bolstering trust models: A key feature of Chainlink today is the ability of users to
select nodes based on decentralized records of their performance histories, as discussed
in Section 3.6.4. The staking mechanism and Implicit-Incentive Framework we intro-
duce in Section 9 together constitute a broadly scoped and rigorous mechanism-design
framework that will empower users with a greatly expanded ability to gauge the secu-
rity of DONs. This same framework will also make it possible for DONs themselves
to enforce various security requirements on participating nodes and ensure operation
within strong trust models.
It is also possible using tools described in this paper for DONs to enforce spe-
cial trust-model requirements, such as compliance with regulatory requirements. For
example, using techniques discussed in Section 4.3, nodes can present evidence of
node-operator characteristics, e.g., territory of operation, that can be used to help
enforce compliance with, e.g., the General Data Protection Regulation (GDPR) Ar-
ticle 3 (“Territorial Scope”) [105]. Such compliance can otherwise be challenging to
meet in decentralized systems [45].
Additionally, in Section 7 we discuss plans to strengthen the robustness of DONs
through trust-minimization mechanisms on the main chains they support.
3 Decentralized Oracle Network Interface and Ca-
pabilities
Here we briefly sketch the capabilities of DONs in terms of the simple but powerful
interface they are designed to realize.
Applications on a DON are composed of executables and adapters. An executable is
a program whose core logic is a deterministic program, analogous to a smart contract.
An executable also has a number of accompanying initiators, programs that call entry
points in the executable’s logic when predetermined events occur—e.g., at certain times
(like a cron job), when a price crosses a threshold, etc.—much like Keepers (see Sec-
tion 3.6.3). Adapters provide interfaces to off-chain resources and may be called by
either the initiators or core logic in executables. As their behavior may depend on that
of external resources, initiators and adapters may behave non-deterministically.
We describe the DON developer interface and the functioning of executables and
adapters in terms of the three resources typically used to characterize computing sys-
tems: networking, compute, and storage. We give a brief overview of each of these
resources below and provide more details in Appendix B.
23
3.1 Networking
Adapters are interfaces through which executables running on a DON can send and
receive data from off-DON systems. Adapters may be viewed as a generalization of
the adapters used in Chainlink today [20]. Adapters may be bidirectional—i.e., they
cannot just pull, but push data from a DON to a web server. They may also leverage
distributed protocols as well as cryptographic functionality such as secure multi-party
computation.
Figure 9: Adapters connecting a DON, denoted DON1, with a range of different re-
sources, including another DON, denoted DON2, a blockchain (main chain) and its
mempool, external storage, a web server, and IoT devices (via a web server).
Examples of external resources for which adapters might be created are shown
in Fig. 9. They include:
• Blockchains: An adapter can define how to send transactions to a blockchain and
how to read blocks, individual transactions, or other state from it. An adapter
can also be defined for a blockchain’s mempool. (See Section 3.5.)
• Web servers: Adapters can define APIs through which data may be retrieved
from web servers, including legacy systems that are not specially adapted for
interfacing with DONs. Such adapters can also include APIs to send data to
such servers. The web servers to which a DON connects may serve as gateways
to additional resources, such as Internet-of-Things (IoT) devices.
24
• External storage: An adapter can define methods to read and write to storage
services outside the DON, such as a decentralized file system [40, 188] or cloud
storage.
• Other DONs: Adapters can retrieve and transmit data between DONs.
We expect that initial deployments of DONs will include a set of building block
adapters for such commonly used external resources and will further allow DON-specific
adapters to be published by DON nodes. As smart contract developers write adapters
today, we expect that they will build even more powerful adapters using this advanced
functionality.
We expect that ultimately it will be possible for users to create new adapters in a
permissionless manner.
Some adapters must be constructed in a way that ensures the persistence and avail-
ability of external resources controlled by a DON. For example, cloud storage may
require maintenance of a cloud services account. Additionally, a DON can perform
decentralized management of private keys on behalf of users (as in, e.g., [160]) and/or
executables. Consequently, the DON is capable of controlling resources, such as cryp-
tocurrency, that may be used, e.g., for sending transactions on a target blockchain.
See Appendix B.1 for further details on DON adapters, as Appendix C for a few
example adapters.
3.2 Computation
An executable is the basic unit of code on a DON. An executable is a pair exec =
(logic, init). Here, logic is a deterministic program with a number of designated entry
points (logic1, logic2,..., logic𝓁) and init is a set of corresponding initiators
(init1, init2,..., inite). To ensure the full auditability of the DON, an executable’s logic
uses the underlying ledgerL for all inputs and outputs. Thus, for instance, any adapter
data serving as input to an executable must be stored first on L.
Initiators: Initiators in Chainlink today cause event-dependent job executions on
Chainlink nodes [21]. Initiators in DONs function in much the same way. A DON ini-
tiator, however, is specifically associated with an executable. An initiator may depend
on an external event or state , on the current time, or on a predicate on DON state .
With their dependency on events, initiators may of course behavenon-deterministically
(as of course may adapters). An initiator can execute within individual DON nodes
and so need not rely on an adapter. (See Example 1 below.)
Initiators are an important feature distinguishing executables from smart contracts.
Because an executable can run in response to an initiator, it can effectively operate
autonomously, as of course by extension can a hybrid contract incorporating the exe-
cutable. One form of initiators today are Chainlink Keepers, which provide transaction
25
automation services, triggering smart contract execution—such as liquidation of un-
dercollateralized loans and execution of limit-order trades—based on oracle reports.
Conveniently, initiators in DONs may also be viewed as a way of specifying the
service agreements that apply to an executable, as they define the circumstances under
which the DON must call it.
The following example illustrates how initiators work within an executable:
Example 1 (Deviation-triggered price feed) . A smart contract SC may require fresh
price-feed data (see Section 3.6.3) whenever there is a substantial change, e.g., 1%, in
the exchange rate between a pair of assets, e.g., ETH-USD. Volatility-sensitive price
feeds are supported in Chainlink today, but it is instructive to see how they can be
realized on a DON by means of an executable execfeed.
The executable execfeed maintains the most recent ETH-USD price r onL, in the
form of a sequence of ⟨NewPrice : j,r⟩ entries, where j is an index incremented with
each price update.
An initiator init1 causes each nodeOi to monitor the current ETH-USD price for
deviations of at least 1% from the most recently stored price r with index j. Upon
detection of such a deviation, Oi writes its current view ri of the new price to L using
an entry of the form ⟨PriceView :i,j + 1,r i⟩.
A second initiator init2 fires when at least k such PriceView-entries with new price
values for index j + 1 created by distinct nodes have accumu프로젝트로 돌아가기