owned this note
owned this note
Published
Linked with GitHub
---
tags: Time, Ethereum2.0
---
# Why clock sync matters in Ethereum 2.0
**Abstract**
Beacon chain protocol critically depends on clock synchronization. If validators or nodes clocks discrepancy becomes too much, then [liveness properties](https://en.wikipedia.org/wiki/Liveness) can become violated, i.e. the protocol participants won't be able to make progress or its probability becomes too low. In theory, [safety properties](https://en.wikipedia.org/wiki/Safety_property) can become violated too, since validators which failed to participate due to clock discrepancies are penalized. If an adversary manages to attack otherwise honest validator clocks then it can seize voting power over time.
## Safety, liveness and robustness
[Casper FFG](https://arxiv.org/abs/1710.09437) is based on epoch justification and finalization. An epoch justification requires 2/3 of total voting power, so if an adversary gains control on 2/3 then it will reign over a Casper FFG based system.
If an adversary controls less than 2/3 of voting power, but it controls 1/3, it might be able to hinder protocol progress, by voting differently than other validators, thus breaking liveness property.
However, even if an adversary controls less than 1/3 of voting power, there are other sources of failures in real network. E.g. there are network delays, clock disparities, hardware or link failures. All these result in additional faults, which combined with an adversarial behavior of some nodes can lead to situations when epochs cannot get justified for some periods of time.
So, from a practical point of view, while protocol can successfully withstand safety and liveness attacks, it can do that with degrade performance, which is a real problem for real users of the protocol.
While, in theoretical sense, liveness property cannot be violated in a finite execution of a protocol, a practical liveness as perceived by real users *can* be violated.
An illustration of the problem can be found in Vitalik Buterin's [post](https://ethresear.ch/t/network-adjusted-timestamps/4187), which also outlines potential solution.
Thus, for designers of practical BFT systems, it's important to design Robust BFT protocols, e.g. able to withstand attacks with relatively low performance degradation.
## Clock disparity consequences
The focus of the document is faults - and attacks - related to lack of clock synchronization.
Beacon chain specs require either delay or discard messages (blocks or attestations), which arrived beyond admissible time bounds. So if a validator is severely out of time, then it's a form of failure. A late message arrival can be a consequence of different problems:
- slow or faulty links
- clock skew or drift
- slow or faulty validators
- slow or faulty transient nodes (in p2p network)
A message can arrive before designated time, which is a problem with clocks either on a sender (fast) or on a receiver (slow) side. However, in the case it will typically be enqueued until the target period of time begins. However, fast clock of a sender (relative to others) can still be a problem, since it will send a message before it gathers prior (in terms of logical time) protocol messages from other nodes. Thus, in certain cases, it can result in a beacon state view, that is different from others. Thus, it also can be seen as a form of fault, which hinders reaching consensus. In other words, it reduces probability that honest validators reach 2/3 majority, aiming to an epoch. Such faults can be summed up with the adversary voting power, alleviating practical liveness or robustness attacks.
### Correctness
As validators are penalized when they evade their duties, timing faults can lead to a situation where otherwise honest validators are gradually loosing their balances. Thus certain attacks are imaginable, where voting power of honest validators degrades while malicious voting power stays the same (or degrades at slower rate). So, an adversary can reach voting power dominance over time, leading to safety property violation.
While it's unlikely in practice (because of counter-measures undertaken by administrators), in theory it's possible that time-level attacks can be used to violate safety properties of the beacon chain protocol (while Caper FFG's *accountable safety* is not viloated).
### Problem persistence
One serious problem is that clock disparity among nodes will grow with time, e.g. if it reaches some dangerous level, it will likely persist until corrected. That is caused by the fact that individual clock rates are not stable and drift because of various reasons, including temperature variations, aging, radiation, etc.
This is in contrast to message transmission delays, which has similar impact, but periods of higher than usual network delays do not typically last for a long.
However the persistence of clock disparity simplifies problem detection.
Thus, a reliable clock synchronization mechanism is a must-level requirement.
### Disparity level
Different level of clock disparity affect protocol in different ways. An attestation can be included in a block, during 32 slots, i.e. the same duration as the duration of an epoch. So, from attestation processing point of view, only severe disparity will critically affect beacon chain protocol (as also illustrated by the Vitalik's [post](https://ethresear.ch/t/network-adjusted-timestamps/4187)).
However, a reward for an attester will be lower, if an attestation is included later. Additionally, delayed inclusion may lead to a delayed fork switch, which is also not good from stability point of view.
Clock disparity are much more severe for block producers. If a block proposer is 1/3 slot late, then attesters won't receive it at the moment when they should attest current head. However, they still can attest the parent, so if the late proposed block references the parent as well, it's not a problem, since the newly proposed block is the only child of the parent block and will be chosen as a head block, when it has arrived.
But if the block proposer is one slot late, then the next slot proposer won't receive the block at time, won't be able to choose it as head and will create its block based on some other block as parent. So, there will be a fork. While beacon chain protocol is able to tolerate forks, together with other problems it can lead to problems, so it's better to reduce such opportunities as much as possible.
Being too fast also can lead to problems. If an attester votes too early, then it won't receive latest information which will be received by other nodes, so its view on beacon chain will be different, which may hinder consensus at some point. However, if there are no forks it shouldn't be a problem, since it still votes for the same chain. That's why it's worse situation when there are forks.
If a block proposer is too fast (about a slot) it won't receive a block from the previous slot proposer, which means a fork happen.
So, clock disparity about 1/3 of a slot affects normal message flow, though should not lead to big problems (unless there are other problems).
Clock disparity around slot duration leads to forks, which is a prerequisite for other problems. Clock disparity around epoch duration are very severe.
However, a recent change to [p2p-interface spec](https://github.com/ethereum/eth2.0-specs/blob/dev/specs/phase0/p2p-interface.md#configuration) assumes maximum clock disparity to be 500ms and prescribes to delay too early blocks (the similar behavior prescribed to attestation processing, but they can be included during a longer time, e.g. during 32 slots, so it's less of a problem). That means even a transient node with even slight clock disparity (fast more than 500 ms) will delay block flow through it.
So, in general, the requirements regarding clock disparity are rather strict, it should be around 100-200ms, since there is also message latency about the same magnitude.
## Time-level attacks
As clock disparity can be a source of faults, at least degrading protocol performance (or leading to liveness/safety violations, in general, as the magnitude of disparity grows), one should investigate ways, how such attack could be performed and harden the system, if needed.
As individual node clocks drift over time, the clock disparity inevitably grows, so clocks should be synchronized periodically. Thus, synchronization protocol is one of the targets that can be attacked.
Unfortunately, in the context of the Internet, NTP is the default (cheap and convenient) way to synchronize clocks, while NTP lacking BFT properties. So, if an adversary can find ways to become an NTP server of choice for a significant number of otherwise honest nodes, then it can manipulate their time, gradually hastening or slowing their clocks and induce timing faults.
As pointed out earlier, even slow discrepancies, like several seconds can lead to timing faults, while can be difficult to detect by unaware administrators (unless the problem is addressed explicitly).
### NTP-level attacks
NTP servers are known to be [misused and abused](https://en.wikipedia.org/wiki/NTP_server_misuse_and_abuse). NTP server bandwidth is a scare resource worldwide. As a consequence, NTP server access becomes more structured and administered.
One initiative is [NTP pool](https://www.pool.ntp.org/en/), which is a big cluster of community time servers and `the default "time server" for most of the major Linux distributions and many networked appliances`.
Therefore, it's likely that many (or even most) validator nodes will use the NTP pool as their time server.
On the other side, joining the pool would not be a problem for an adversary
> Because of the large number of users we are in need of more servers. If you have a server with a static IP address always available on the internet, please consider adding it to the system.
While NTP is able to tolerate certain faults, including "falsetickers", it's limited in the ability. I'm not an expert in NTP and the following reasoning needs revision, however, it's likely that such an adversary has a high chance to be able to manipulate the NTP pool clients, if it controls significant portion of the pool.
So, if the adversary adds about 4000 own time servers to the pool, they will comprise half of the pool. So, there will be 1/2 chance that a time server chosen randomly from the pool is controlled by the adversary.
Typical NTP configuration includes 4 NTP servers. So, if there is only one malicious time server, it will be recognized as a "falseticker". However, if there are two malicious and two regular servers, then NTP is not able to recognize malicious servers as "falsetickers". Moreover, as two servers may report extremely correlated time readings, they are likely to be preferred.
So, given 0.5 probability to choose a malicious server from the pool, the are three outcomes of choosing 4 of them:
- honest time servers dominate, i.e. 3 or 4 of them, this is 1/16 + 4/16 = 5/16 probability
- malicious time servers dominate, the same 5/16 probability
- two honest against two malicious time servers, 6/16 probability. It's not clear who will win but it's a risky setup, since either two malicious server will look as more "synchronous" or they will affect average reading of the time server quorum (the exact NTP implementation behavior should be additionally investigated)
Overall, by "donating" around 4K time servers, an adversary has about 1/2 chance to affect time on a node, which is configured to use NTP pool. The chance can probably be increased, by exploiting details of NTP and/or NTP pool administration procedures.
Note, that such an adversary time pool can behave honestly to a regular node, if its IP is not in `validators` list. It can also behave correctly for some period of time for IPs in the list and start to manipulate clocks later. It can also detect that it's the single malicious timeserver for a particular node (high chance to be identified as a "falseticker") and service it normal way. From BFT literature perspective, Byzantine faulty clock can return different values when read by different processes and this is the property, which makes BF tolerant clock sync protocol difficult/expensive.
## Mitigating time-level attacks
The main purpose of the document is to describe the problem and to give foundation to the thesis *Why Clock Synchronization matters in Ethereum 2.0*. So, I will only briefly discuss how the clock synchronization problem can be solved. A more detailed solution is to be described in a separate document.
### Appropriate NTP configuration
The potential NTP pool attack problem described above can be solved by an appropriate configuration of NTP protocol - do not use the default configuration, but choose other NTP servers instead. However, a problem that remains is which NTP servers should be used instead.
Taking an NTP server from a publicly available list can expose to the same kind of problems, if an adversary manages to add its servers to such a list.
One can use NTP servers operated by government agencies, well known commercial or non profit organization. However, it means the presumably decentralized PoS architecture critically depends on relatively small list of time servers, so it becomes a kind of Proof-of-Authority in some sense.
While it might be acceptable (the protocol relies on IP, DNS, etc anyway), there are risks associated with this.
Servers can cease servicing or can participate in clock manipulation.
So, a truly decentralized BFT system deserves more BF tolerant clock synchronization protocol. At least, such solutions should be proposed and analyzed.
### Anonymizing validator nodes
As written before, Byzatine clock return different results, when read by different processes. A publicly available, but Byzantine NTP server (e.g., included in some public NTP server list) should service correctly normal users, else it will be easily identified as faulty.
Thus such janus-faced server should be able to discriminate validator nodes from other clients in order to manipulate former.
So, a simple intermediate server can conceal a validator node, especially if it uses two IP addresses (to service the validator node requests from an IP which is distinct from the IP used to query an NTP server).
However, such a setup increases costs of being a validator, so it will be a problem to many low-steak validators.
### GNSS or radio clocks
A similar approach is to use other publicly available time services, e.g. [GNSS](https://en.wikipedia.org/wiki/Satellite_navigation) receivers or [radio clock](https://en.wikipedia.org/wiki/Radio_clock). Inexpensive GPS receivers are available and such service access by validator nodes cannot be discriminated from others, so such services cannot be used to manipulate validator clocks.
The same concerns about centralization and a higher entry barriers apply here. However, the former one doesn't seem to be a really serious one - people and organization over the world rely on international standards including UTC time standard. It's inevitable in some sense, including the requirement that beacon chain slot duration should be equal to `SECONDS_per_SLOT`, which also relies on international time standard.
So, it could be considered as almost perfect solution with the exception that it raises an entry barrier for some validators, e.g. if one would like to use server hosting to run a validator. However, they can set up an alternative clock synchronization scheme.
### BFT Clock Synchronization
From BF tolerance perspective, as beacon chain critically relies on Clock Synchronization property, there should be BFT Clock Synchronization (sub)protocol.
The NTP pool attack above illustrates one of the main problems of practical BFT solutions: one has to restrict participants of a BFT system by some means, else a Sybil-like attack is possible, i.e. an adversary can add many participants which will surpass safety limits (like a typical 'less than 1/3 of faulty participants' or so).
NTP pool is an open (to join) system, which makes it a problem to rely on it in a BFT system.
A straightforward way to restrict participants is to require that (only) validators (can) participate in creating a BFT service. Vitalik Buterin also discusses this in the [post](https://ethresear.ch/t/network-adjusted-timestamps/4187) mentioned above.
A number of BFT clock synchronization protocol exists, however, the cost of such protocols can be prohibitive, given the expected amount of Ethereum 2.0 validator/nodes (an order of thousand of nodes).
As p2p architecture is used to connect the validator nodes, validators cannot reach each other directly, thus message delays are multiplied by some factor (an average/maximum number of hops), which limits clock synchronization accuracy, because network delay variability decreases the ability to measure clock offsets. However, an estimation of such delays based on special (asymmetric) optimization measures can be used (e.g. https://www.researchgate.net/publication/3958812_Clock_synchronization_algorithms_for_network_measurements).
### Redesigning beacon chain protocol
An alternative approach would be to re-design beacon chain to avoid its reliance on the clock synchronization property. While it seems quite radical, actually it's a traditional approach to build reliable distributed systems, since logical clocks are much easier to implement, while implementing BFT clock synchronization with beacon chain level of requirements together with the expected scale of the system (thousands of nodes) can be quite challenging.
However, cryptoeconomics still require that slot duration be of the prescribed amount (`SECONDS_PER_SLOT`, 12 seconds currently), because reward rate depends on slot/epoch rate. Thus clock synchronization with the world time is still required. However, clock disparity requirements are mild relative clock disparity required to align validator slots along the same time axis. Because, reward rate should be stable on a long scale.
However, EVM or smart contracts may require higher quality of clock synchronization.