Engineering Feature Show

2025 Q1

New in Aiken

new in Aiken

Faster Execution.

  • cpu: from -4% to -24%
  • mem: from -7% to -26%

LP Mint :: Mint Two

Swap A for B

NFT Marketplace: buy one

new in Aiken

New builtins.

  • and_bytearray(padded: Bool, left: ByteArray, right: ByteArray) -> ByteArray
  • or_bytearray(padding: Bool, left: ByteArray, right: ByteArray) -> ByteArray
  • xor_bytearray(padding: Bool, left: ByteArray, right: ByteArray) -> ByteArray
  • complement_bytearray(self: ByteArray) -> ByteArray
  • rotate_bytearray(self: ByteArray, offset: Int) -> ByteArray
  • shift_bytearray(self: ByteArray, offset: Int) -> ByteArray
  • count_set_bits(self: ByteArray) -> Int
  • find_first_set_bit(self: ByteArray) -> Int
  • read_bit(self: ByteArray, index: Int) -> Bool
  • write_bits(self: ByteArray, indices: List<Int>, value: Bool) -> ByteArray
  • replicate_byte(length: Int, byte: Int) -> ByteArray
  • ripemd_160(preimage: ByteArray) -> ByteArray
  • unconstr_fields(data: Data) -> List<Data>
  • unconstr_index(data: Data) -> Int

https://aiken-lang.github.io/prelude/aiken/builtin.html

new in Aiken

Maths in docs.

new in Aiken

Benchmarks.

new in Aiken

Types as Namespaces

new in Aiken

Bug fixes & QoL improvements.

  • Fix aiken blueprint policy computing hashes as PlutusV1, instead of relying on the plutus version from the Blueprint.
  • Fixed a code gen crash when using records in when is expressions.
  • Panic error when using aiken uplc decode on cbor encoded flat bytes.
  • Comment formatting in pipelines leading to confusion.
  • Preserve holes discard name in function captures. See #1080.
  • Disambiguate type-alias blueprint definition using module's name. See #1074.
  • Fix blueprint schema definitions related to Pairs definitions, and generate them as data List.
  • Fixed UnknownTypeConstructor wrongly reported as UnknownVariable (then messing up with LSP quickfix suggestions).
  • Formatter was removing comments from function type annotation args.
  • Parser wrongly merged two adjacent sequences together, effectively fusioning scopes.
  • Fix hint when suggesting to use named fields, wrongly suggesting args in lexicographical order instead of definition order.
  • Fix aiken blueprint apply wrongly overriding all validators handlers names & ABI to the mint's one. See #1099.
  • Prevent (type error) backpassing blocks with empty continuation. See #1111.
  • Change default placeholder for trace to Void instead of todo.
  • Disallow (parse error) dangling colon : in traces. See #1113.
  • Additional code action to use constructors or identifiers from qualified imports is now offered on missing constructor or identifier.

  • eval_phase_two and related functions now return an EvalResult.

  • New -S flag on check and build that blocks the printing of warnings but it still shows the total warning count.
  • Fix aiken blueprint policy computing hashes as PlutusV1, instead of relying on the plutus version from the Blueprint.
  • Fixed a code gen crash when using records in when is expressions.
  • Panic error when using aiken uplc decode on cbor encoded flat bytes.
  • Comment formatting in pipelines leading to confusion.
  • Preserve holes discard name in function captures. See #1080.
  • Disambiguate type-alias blueprint definition using module's name. See #1074.
  • Fix blueprint schema definitions related to Pairs definitions, and generate them as data List.
  • Fixed UnknownTypeConstructor wrongly reported as UnknownVariable (then messing up with LSP quickfix suggestions).
  • Formatter was removing comments from function type annotation args.
  • Parser wrongly merged two adjacent sequences together, effectively fusioning scopes.
  • Fix hint when suggesting to use named fields, wrongly suggesting args in lexicographical order instead of definition order.
  • Fix aiken blueprint apply wrongly overriding all validators handlers names & ABI to the mint's one. See #1099.
  • Prevent (type error) backpassing blocks with empty continuation. See #1111.
  • Change default placeholder for trace to Void instead of todo.
  • Disallow (parse error) dangling colon : in traces. See #1113.
  • Additional code action to use constructors or identifiers from qualified imports is now offered on missing constructor or identifier.

  • eval_phase_two and related functions now return an EvalResult.

  • New -S flag on check and build that blocks the printing of warnings but it still shows the total warning count.

Amaru

Terminology

Node Type Fetch data & submit txs from peers Validate data (i.e. do not trust peers) Participate in txs & block propagation Can forge & produce blocks Can validate historical data
Data ✅
 ❌ ❌ ❌ ❌
Proxy ✅ ✅ ❌ ❌ ❌
Relay ✅ ✅ ✅ ❌ ❌
Full ✅ ✅ ✅ ✅ ❌
Archive ✅ ✅ ✅ ✅ ✅

The Haskell Node

Why build another node?

Good

Bad

Ugly

Exists

Reasonably stable

Designed end-to-end

Well-documented

Archive-capable

Controlled by one entity

Unmaintainable parts

Mixing responsibilities

Hard to radically extend

Heavy resource usage

Limited monitoring

Limited/pedantic interfaces

Developed end-to-end

Challenges

Amaru's challenges

Peer management

  • Random connections to arbitrary peers.
    • May or may not send valid information
    • May or may not abuse our own resources
    • May or may not be a sibyl actor
  • Distributed system
    • Messages (i.e. blocks) typically arrive out-of-order
    • Information takes time to propagate
  • PoS & block legitimacy
    • Requires intrinsic knowledge (stake distribution)
    • Private leader-schedule
    • Blocks are computationally cheap to forge

  1. Distributed

  2. Unknown peers

  3. No extrinsic truth

Amaru's challenges

Peer management

  1. Distributed

  2. Unknown peers

  3. No extrinsic truth

Amaru's challenges

Multiple realities

How people typically view "the blockchain"

How it actually is

up to 2160 levels

  • Stochastic settlement
    • In practice, the probability of long-fork exponentially decreases with length, but increases with adversarial stake
    • Attacks still feasible with <50% of the stake
       
  • Potentially as many chains as there are peers
    • Must track all plausible forks
    • Must remember who sent what (peer management)
  • Slot battles
    • Induced by the private-leader schedule
    • Cause regular small forks/rollbacks

  1. Private schedule

  2. Stochastic

  3. No single truth

Amaru's challenges

Multiple realities

How people typically view "the blockchain"

How it actually is

up to 2160 levels

  1. Private schedule

  2. Stochastic

  3. No single truth

Amaru's challenges

Δ < 1s

  1. Block propagation

  2. Slow I/O

  3. Binomial distribution

  • Block propagation must* reach 95% of network within 5s
    • *For optimal network operation;
    • Can still reach consensus otherwise but with degraded performances.
    • Latency worldwide ~400-500ms
  • In practice, any block validation must be done within 1s
    • Binomial distribution of blocks due to random leader schedule
      • 1 block every 20s on average
      • 50% of blocks between 1s and 14s
      • Significant portion (5%) at 1s
         
  • I/O is "slow", but required for persistence
    • Trade-offs between RAM vs commodity storage
    • Optimized for worse-case (Byzantine environment)

Amaru's challenges

Δ < 1s

  1. Block propagation

  2. Slow I/O

  3. Binomial distribution

Amaru's challenges

Deferred computation

e
e+1
e+2
e+3

delegation

snapshot(e+1)

snapshot(e+2)

snapshot(e+3)

snapshot (e)

leader schedule

computed from e

computing rewards from:

  • stake distr at e
  • pool performances at e+2

distributing rewards

from e computed in e+3

Amaru's challenges

Deferred computation

  • Rewards are calculated using past stake distributions
    • Need data from various epochs
    • Mainnet: ~1.5M accounts, ~11M UTxO, ~3000 pools
    • Paid directly into accounts, no on-chain materialization
    • Influences leader-schedule
  • Randomness is obtained from past epochs (hence Ouroboros)
    • VRF nonces made of randomness commitments collected during past epochs.
    • Combined with other data from even earlier epochs.
  • Governance uses live stake distributions
    • For both stake pools and DReps, live stake is used
    • Computing stake distribution is long if stored on disk
    • Storing stake distribution in memory increase resource usage

  1. Rewards

  2. Cyclic randomness

  3. Governance

e
e+1
e+2
e+3

snapshot (e)

delegation

snapshot(e+1)

snapshot(e+2)

snapshot(e+3)

leader schedule

computed from e

computing rewards from:

  • stake distr at e
  • pool performances at e+2

distributing rewards

from e computed in e+3

Amaru's challenges

Deferred computation

  1. Rewards

  2. Cyclic randomness

  3. Governance

Engineering Feature Show - 2025 Q1

By Matthias Benkort

Made with Slides.com

Engineering Feature Show - 2025 Q1

  • 121

More from Matthias Benkort