This document provides technical details on Azimuth's "Layer 2" scaling solution
for Azimuth, known more formally as "naive rollups". We focus here primarily on the
"Hoon smart contract" located at /lib/naive.hoon in your ship's pier, as well
as other proximal topics.
This is intended for developers that desire a deeper understanding of how this protocol works, how secure it is, and how to extract data from layer 2 to obtain a complete picture of the Arvo network.
This is not intended for everyday users who only wish to know how to either transfer their ship to layer 2 or perform layer 2 actions. This is a functionality of Bridge for which documentation will soon be available. For a casual overview of the rationale and functionality of layer 2, please see this blog post. For more information on how Azimuth works more generally, including interactions with Bridge and Ethereum, see the page on Azimuth data flow.
This page is also not where to find instruction on how to run your own "aggregator"/"roller". Documentation for this process is found here. However, this page does contain essential background information for anybody in this category.
Summary
Naive rollups were developed in response to rising gas costs for performing Azimuth actions. They also serve the dual purpose of making onboarding easier, as it is now possible to acquire a planet and get on Urbit without any knowledge of Ethereum or cryptocurrency.
In this section we give a high-level summary of how naive rollups function and how they affect the end user. Later sections elaborate on this summary.
Layer 1
We briefly review how "Layer 1", i.e. the Azimuth smart contract suite, functions. An update to the Azimuth PKI data stored on your urbit occurs with four steps:
- A transaction is posted to the Ethereum blockchain.
- The Ethereum Virtual Machine calculates the resulting state transition and checks its validity, then updates the state if it is a valid transition.
- Your urbit downloads the new state from an Ethereum node.
- Your urbit makes the final decision on whether the new state is valid.
By default, step four always succeeds. It has always been possible in theory for your urbit to dispute what it read on Ethereum, but there has never been any reason to do so.
Layer 2
Layer 1 still functions identically today as it did before naive rollups. Naive rollups work via the following process.
- A batch of one or more transactions is posted to the Ethereum blockchain by an urbit node called a roller or aggregator.
- Your urbit downloads the transactions from an Ethereum node.
- Your urbit computes the resulting state transitions from the transactions and checks them for validity.
- Your urbit updates its locally stored Azimuth state using state transitions from the batch that have been deemed valid.
In comparison with Layer 1, the EVM no longer checks the validity or computes
the state transitions for any given transaction. It is now being used solely as
a database of submitted transactions, and the business logic of computing what
these transactions mean has been offloaded to your urbit. Thus we think of
naive.hoon as being the first "Hoon smart contract". You could also consider
steps 3 and 4 of the Layer 2 process as being a fattening of the trivial step 4
of the layer 1 process.
Here we briefly elaborate on the layer 2 steps, but see below for more technical detail.
A roller is any urbit node - even a moon, comet, or fakezod will do - to which batches of transactions are submitted. You could use your own ship as a roller if you wanted. The roller collects batches of transactions from whichever parties they choose and submits them to the Ethereum blockchain all at once. We expect this to either happen on a regular interval, or once some minimum threshold of transactions is reached, but the decision on when to submit is ultimately up to the roller.
Computing the resulting state transitions obtained from downloaded Ethereum
transactions is done using /lib/naive.hoon, which is a gate that accepts both layer
1 transaction logs and layer 2 batches, and then computes the resulting state
transitions and updates the ship's internal Azimuth data held in
the Gall agent /app/azimuth.hoon accordingly.
Savings in gas costs
There are several dimensions by which naive rollups saves on gas over layer 1. They are:
- Gas is not spent on instructing the EVM to compute state transitions and confirm validity.
- Layer 2 Azimuth state is not stored on Ethereum, so the only data storage gas costs is for the transactions themselves.
- Layer 2 transactions are written in a highly compressed form. E.g., instead
of calling the spawn action by name, it is simply referred to as action
%1. - By collecting multiple layer 2 transactions and submitting them as a single "batch transaction", additional gas savings are achieved by not needing to duplicate information such as which smart contract the transactions are intended for.
Put together, these create a reduction in gas costs of at least 65x when adding a transaction to a sufficiently large batch (approximately 30 or more transactions). A single transaction submitted as a batch is approximately 5x cheaper, while 10 transaction submitted as a batch is approximately 30x cheaper. Thus using one's own ship to submit a single-transaction batch is still a cost-saving measure.
One way trip
Moving to layer 2 is a one-way trip for now. This means that once a ship moves to layer 2, it cannot be moved back to layer 1. We believe it to be technically possible to engineer a return trip, and expect that someday this will be the case, but there are no plans to implement this in the near future.
Interacting with L2
Layer 2 ships can perform the same actions on layer 2 that they could on layer 1, but are no longer able to perform any layer 1 actions. Layer 1 ships can also perform a subset of layer 2 actions - namely the ones related to sponsorship.
For a complete list of what layer 2 actions each ship rank, layer, and proxy can perform, see Layer 2 Actions. For an explicit description of the byte format of Layer 2 Ethereum transaction, see Bytestring format.
Sponsorship
Due to the possibility of sponsors and sponsees existing on different layers, the precise logic of how sponsorship works is complex. However, under common circumstances it is simple.
If either the sponsor or sponsee are on layer 2, then sponsorship actions must occur on layer 2. The only exception to this is detaching. A sponsor on layer 1 may perform a layer 1 detach action on a layer 2 sponsee, and this will result in the sponsee having no sponsor on layer 1, and layer 2 as well if they were the sponsor on layer 2. This is necessary to simplify the logic, but it also guarantees that there is no hard requirement to ever utilize layer 2. Without this exception, sponsors with sponsees that move to layer 2 would be forced to detach them as a layer 2 action if they wanted to cease sponsorship. This would also have an unacceptable impact on ships owned by smart contracts.
If both sponsor and sponsee are on layer 1 then sponsorship actions may occur on either layer. As long as all sponsorship actions betweeen the two parties occur on a single layer, behavior will be as expected.
In most cases this is sufficient to understand how sponsorship works. However there are a number of edge cases that make this more complicated that developers may need to concern themselves with in scenarios where layer 1 sponsor and sponsees are mixing layer 1 and layer 2 actions. In the Sponsorship state transitions section below, we give a table that shows how the sponsor and escape status of a ship changes according to which actions are taken.
Smart contracts
Smart contracts are unable to own layer 2 ships, and thus cannot sign layer 2 transactions. This creates a hard requirement that layer 1 ships be allowed to perform a layer 1 detach operation on a layer 2 ships.
Azimuth state
The introduction of layer 2 presents additional complication in understanding Azimuth state. In order to be precise we define the following terminology:
- Layer 1 Azimuth state refers to the state of Azimuth as reflected on the Ethereum blockchain. This excludes all layer 2 transactions. Depositing to layer 2 is considered a layer 1 action, so the layer 1 Azimuth state is aware of which ships are on layer 2, but is blind to everything that happens to them afterward.
- Layer 2 Azimuth state refers to the state of Azimuth as stored in
/app/azimuth.hoonon your ship. The state here takes into account transactions that occur on both layers. No distinction between the layers is made in the state here - e.g. a ship only has one sponsor in Layer 2 Azimuth state, not a layer 1 sponsor and a layer 2 sponsor. This is the state actually in use by Urbit. Layer 1 Azimuth state is now only an input for generating Layer 2 Azimuth state, so any time a ship needs to check e.g. the public key of a ship (regardless of which layer it is on), it will check the Layer 2 Azimuth state, not the Layer 1 Azimuth state. - Layer-2-Only Azimuth state refers to the state of Azimuth as reflected solely by layer 2 transactions. This state is not explicitly stored anywhere, but is computed as part of the process to create the Layer 2 Azimuth state. We do not make any further references to this state, but it is important to keep in mind conceptually.
Layer 1 Azimuth state is computed by the Ethereum Virtual Machine. Layer 2
Azimuth state is computed by taking in the Layer 1 Azimuth state and modifying
it according to layer 2 transactions using /lib/naive.hoon. When we are being
precise about which state we are referring to we will utilize capitalization as above.
Layer 2 Azimuth state is held by the Azimuth Gall app located at
/lib/azimuth.hoon. Layer 1 and layer 2 state are not held separately - your
ship holds only one canonical Azimuth state, generated by parsing both layer 1
and layer 2 Ethereum transactions using /lib/naive.hoon. It is important to
keep in mind that Layer 1 Azimuth state is entirely unaware of Layer 2 Azimuth
state. Thus, for instance, the Azimuth PKI on Ethereum (Layer 1 Azimuth state)
may claim that the sponsor of ~sampel-palnet is ~marzod, while the Azimuth
state held on your ship (Layer 2 Azimuth state) claims that the sponsor of
~sampel-palnet is ~dopzod. Under this circumstance, this would mean that the
sponsor of ~sampel-palnet was ~marzod before ~sampel-palnet was deposited
to layer 2, and thus the Azimuth PKI on Ethereum will forever reflect this.
For more information on how Azimuth state is handled, including how this integrates with Bridge and Ethereum, see Azimuth data flow.
Sponsorship state transitions
When either a sponsor or sponsee is on layer 2, then all sponsorship actions occur on layer 2 and layer 1 Azimuth state is ignored. The exception to this, as noted above, is when a layer 1 sponsor performs a layer 1 detach action on a layer 2 sponsee. Furthermore, any time a ship moves from layer 1 to layer 2, its sponsorship status is automatically maintained in layer 2.
The only potentially complicated scenario is when both sponsee and (potential) sponsor exist on layer 1. Then because layer 1 actions can modify layer 2 state, careful consideration is required for interactions that mix the two. If you and your sponsor/sponsee are not mixing layer 1 and layer 2 sponsorship actions between yourselves, then you have nothing to worry about and may safely ignore this section.
But, for instance, if both ~sampel-palnet and ~dopzod are on layer 1, it is
technically possible for ~sampel-palnet to escape to ~dopzod on layer 1, and
then ~dopzod can accept the escape on layer 2. This will result in ~dopzod
appearing as the sponsor in the Layer 2 Azimuth state (and thus be
~sampel-palnet's "true" sponsor), and the sponsor in the Layer 1 Azimuth state
will remain unchanged. While it is difficult to imagine a good reason to do
this, developers working with layer 2 need to keep in mind these edge cases and
ought to read on.
In the following table, columns E_1 and S_1 represent the escape status and
sponsor of a given ship as reflected by the Layer 1 Azimuth state. Columns E_2
and S_2 represent the escape status and sponsor of a given ship as reflected
in the Layer 2 Azimuth State. The "true" escape status and sponsor of a ship is
always what is listed in the Layer 2 Azimuth state. In other words, at the end
of the day, S_2 is always the sponsor that matters, but layer 1 actions can
affect the values of E_2 and S_2.
A tar * entry represents any value and if an event shows a transition
from * to * that means that value is not altered by the transition. The
transitions marked with !! are prohibited by the layer 1 Azimuth smart
contract and thus never occur. A1 and A2 represent two distinct ships.
Event | E_1 | E_2 | S_1 | S_2 | -> | E_1 | E_2 | S_1 | S_2
L1-escape A1 | * | * | * | * | -> | A1 | A1 | * | *
L1-cancel A1 | ~ | * | * | * | -> !! :: no cancel if not escaping
L1-cancel A1 | A1 | * | * | * | -> | ~ | ~ | * | *
L1-adopt A1 | A1 | * | * | * | -> | ~ | ~ | A1 | A2
L1-adopt A1 | ~ | * | * | * | -> !! :: no adopt if not escaping
L1-adopt A1 | A2 | * | * | * | -> !! :: no adopt if not escaping
L1-detach A1 | * | * | A1 | A1 | -> | * | * | ~ | ~
L1-detach A1 | * | * | A1 | A2 | -> | * | * | ~ | A2
L1-detach A1 | * | * | A1 | ~ | -> | * | * | ~ | ~
L2-escape A1 | * | * | * | * | -> | * | A1 | * | *
L2-cancel A1 | * | * | * | * | -> | * | ~ | * | *
L2-adopt A1 | * | A1 | * | * | -> | * | ~ | * | A1
L2-adopt A1 | * | A2 | * | * | -> | * | A2 | * | *
L2-adopt A1 | * | ~ | * | * | -> | * | ~ | * | *
L2-reject A1 | * | A1 | * | * | -> | * | ~ | * | *
L2-reject A1 | * | A2 | * | * | -> | * | A2 | * | *
L2-reject A1 | * | ~ | * | * | -> | * | ~ | * | *
L2-detach A1 | * | * | * | A1 | -> | * | * | * | ~
L2-detach A1 | * | * | * | A2 | -> | * | * | * | A2
L2-detach A1 | * | * | * | ~ | -> | * | * | * | ~Aggregators
An "aggregator" or "roller" is any Urbit node that collects signed layer 2
transactions (typically via /app/azimuth-rpc.hoon), combines them into a
"batch", and then submits the batch as an Ethereum transaction. Any urbit can be
a roller, including moons, comets, and even fakezods. You can also use your own
ship as a roller.
Tlon has set up our own roller that is free to use by the community. Using Bridge, a ship may submit X transactions to Tlon's roller per Y period free of charge. Tlon's roller submits on a regular schedule: a submission occurs when a total of A layer 2 transactions have been submitted to it since the the previous submitted Ethereum transaction, or every Z time, whichever occurs first.
There are no security risks in utilizing an aggregator. The transactions you
submit to it are signed with your private key, and so if an aggregator alters
them the signature will no longer match and naive.hoon will reject it as an
invalid transaction. The worst an aggregator can do is not submit your transaction.
Multi-keyfiles
As part of the layer 2 upgrade, Tlon has expanded the role of keyfiles. One of our goals with layer 2 was to reduce the amount of friction experienced when getting onto Urbit. The enormous reduction in fees has made a new boot method which allows instantaneous sale of layer 2 planets or stars to be cost effective.
The ideal situation would be for the end user to be able to buy or receive a planet and immediately boot it without having to wait for an aggregator to submit a transaction that spawns the planet. In order to bring about this circumstance, "multi-keyfiles" have been introduced.
Multi-keyfiles are keyfiles used to boot an urbit for first time that contains more than one set of keys, the purpose of which is to initially utilize one set of keys on a temporary basis, and then the other set of keys soon after. This works as follows. A star owner prespawns a number of layer 2 planets ready to be sold at any time for which they posess the initial key. When a user acquires a planet from a star owner, the star owner immediately hands the initial keys to boot the planet to the buyer. The buyer then queues a transaction at a roller to rotate the keys to a new value only known by the buyer (this step is performed automatically by Bridge). The buyer then downloads a keyfile from Bridge containing both keys and uses that to boot their planet for the first time. After the transaction setting the new keys is submitted by the roller to Ethereum, the purchased planet will automatically switch to them once it reads the corresponding transaction from Ethereum.
This process introduces a short period of time in which both the buyer and seller are in possession of the keys of a planet. A malicious seller could theoretically sell a planet to more than one party in this time period. However, they would quickly be found out as two identical ships on the network immediately creates problems, and the seller's reputation would be tarnished. Due to the expense or effort needed to acquire a star, this seems an unlikely scenario as the reward is much less than the cost. Nonetheless, buyers should always make an effort to purchase from a reputable star, as is the case with all transactions in life. If you want to be absolutely sure that you've received the planet, just wait for the batch to be sent and confirmed.
Multi-keyfiles were possible before layer 2, but as the cost of configuring keys was comparable to the cost of buying a planet, they were not practical.
Security measures
In the process of designing naive rollups, we felt it to be of the utmost
importance that there not be any loss in the security of a layer 2 ship over a
layer 1 ship. In this section we outline several relevant facets of Urbit as
well as particular measures that were taken to ensure that naive rollups were
free of bugs and exploits. We think of naive.hoon as being the first "Hoon
smart contract", and thus its functionality needs to be as rock-solid and
guaranteed as the Azimuth Ethereum smart contracts.
Arvo is deterministic
Crucial to the functionality of Ethereum smart contracts is that they work the
same way every time since the Ethereum Virtual Machine is deterministic.
Similarly, as the state of Arvo is evolved via a single pure
function, Arvo is deterministic as
well. This property makes it well-suited for cases where side effects are
unacceptable such as smart contracts, and thus naive.hoon is worthy of the
name "Hoon smart contract".
Restricted standard library
A standard security practice is to reduce the surface area of attack to be as
minimal as possible. One common source of exploits among all programming
languages are issues with the standard library of functions, and this is one factor
that leads to the existence of multiple implementations of standard library functions for
a given programming language. For instance, glibc is the most widely used
standard library for the C programming language, but sheer size gives a large
surface area in which to find exploits. Thus, other standard libraries have been
written such as musl that are much smaller, and some argue to be more secure
at least partially due to fewer lines of code. Hoon is not yet popular enough to have
multiple standard library implementations, but naive.hoon shucks the usual
standard library and so its subject contains only the exact standard library
functions needed for it to function. This library is known as tiny and is
found at /lib/tiny.hoon.
Unit tests
naive.hoon is among the most well-tested software in Urbit. The test suite,
which may be run with -test %/lib/tests/naive ~ from dojo, is larger than any
other test suite both in number of lines of code and number of tests. We believe
branch coverage to be at or very close to 100%
Nonces
In the Layer 2 Azimuth state, each proxy belonging to a given ship (including
the ownership "proxy") has a non-negative integer associated to it called a
"nonce". Each transaction submitted by a given proxy also have a nonce value. If
the current nonce of a proxy in the Layer 2 Azimuth state is n, then only a
transaction from that proxy with a nonce of n+1 will be considered valid.
Otherwise the transaction is discarded. A valid transaction, by which we
mean one in which the nonce and signature are correct, will increment the nonce
of the proxy in the Layer 2 Azimuth state by one once processed by naive.hoon.
Note that "valid transactions" also include ones where the action will fail,
such as a planet attempting to %spawn another planet. For the purposes of
incrementing the nonce, only the nonce and signature matter.
The use of nonces prevents "replay attacks", which is when a malicious party collects valid transactions and attempts to resubmit them in order to perform the action again. Since a valid transaction increments the nonce associated to the proxy that submitted it, it is only ever possible for a given transaction to alter the Layer 2 Azimuth state once.
The use of nonces also eliminates potential issues caused by submitting a
transaction to more than one roller. This might happen if the first roller
submitted to is taking too long for your liking, and you want to try again with
another. If both rollers end up submitting the transaction, only the first one
will succeed, as the second one will be ignored by naive.hoon for having the
wrong nonce.