Weights to u64 + Balances Weights (#5446)

Co-Authored-By: Kian Paimani <5588131+kianenigma@users.noreply.github.com>
This commit is contained in:
Shawn Tabrizi
2020-04-16 10:43:18 +02:00
committed by GitHub
parent 1c7525fd4b
commit 980b635c8d
82 changed files with 636 additions and 428 deletions
+98 -80
View File
@@ -16,9 +16,10 @@
//! Some configurable implementations as associated type for the substrate runtime.
use core::num::NonZeroI128;
use node_primitives::Balance;
use sp_runtime::traits::{Convert, Saturating};
use sp_runtime::{Fixed64, Perbill};
use sp_runtime::{Fixed128, Perquintill};
use frame_support::{traits::{OnUnbalanced, Currency, Get}, weights::Weight};
use crate::{Balances, System, Authorship, MaximumBlockWeight, NegativeImbalance};
@@ -51,8 +52,7 @@ pub struct LinearWeightToFee<C>(sp_std::marker::PhantomData<C>);
impl<C: Get<Balance>> Convert<Weight, Balance> for LinearWeightToFee<C> {
fn convert(w: Weight) -> Balance {
// substrate-node a weight of 10_000 (smallest non-zero weight) to be mapped to 10^7 units of
// fees, hence:
// setting this to zero will disable the weight fee.
let coefficient = C::get();
Balance::from(w).saturating_mul(coefficient)
}
@@ -60,16 +60,16 @@ impl<C: Get<Balance>> Convert<Weight, Balance> for LinearWeightToFee<C> {
/// Update the given multiplier based on the following formula
///
/// diff = (previous_block_weight - target_weight)
/// diff = (previous_block_weight - target_weight)/max_weight
/// v = 0.00004
/// next_weight = weight * (1 + (v . diff) + (v . diff)^2 / 2)
/// next_weight = weight * (1 + (v * diff) + (v * diff)^2 / 2)
///
/// Where `target_weight` must be given as the `Get` implementation of the `T` generic type.
/// https://research.web3.foundation/en/latest/polkadot/Token%20Economics/#relay-chain-transaction-fees
pub struct TargetedFeeAdjustment<T>(sp_std::marker::PhantomData<T>);
impl<T: Get<Perbill>> Convert<Fixed64, Fixed64> for TargetedFeeAdjustment<T> {
fn convert(multiplier: Fixed64) -> Fixed64 {
impl<T: Get<Perquintill>> Convert<Fixed128, Fixed128> for TargetedFeeAdjustment<T> {
fn convert(multiplier: Fixed128) -> Fixed128 {
let block_weight = System::all_extrinsics_weight();
let max_weight = MaximumBlockWeight::get();
let target_weight = (T::get() * max_weight) as u128;
@@ -78,19 +78,20 @@ impl<T: Get<Perbill>> Convert<Fixed64, Fixed64> for TargetedFeeAdjustment<T> {
// determines if the first_term is positive
let positive = block_weight >= target_weight;
let diff_abs = block_weight.max(target_weight) - block_weight.min(target_weight);
// diff is within u32, safe.
let diff = Fixed64::from_rational(diff_abs as i64, max_weight as u64);
// safe, diff_abs cannot exceed u64 and it can always be computed safely even with the lossy
// `Fixed128::from_rational`.
let diff = Fixed128::from_rational(
diff_abs as i128,
NonZeroI128::new(max_weight.max(1) as i128).unwrap(),
);
let diff_squared = diff.saturating_mul(diff);
// 0.00004 = 4/100_000 = 40_000/10^9
let v = Fixed64::from_rational(4, 100_000);
// 0.00004^2 = 16/10^10 ~= 2/10^9. Taking the future /2 into account, then it is just 1
// parts from a billionth.
let v_squared_2 = Fixed64::from_rational(1, 1_000_000_000);
let v = Fixed128::from_rational(4, NonZeroI128::new(100_000).unwrap());
// 0.00004^2 = 16/10^10 Taking the future /2 into account... 8/10^10
let v_squared_2 = Fixed128::from_rational(8, NonZeroI128::new(10_000_000_000).unwrap());
let first_term = v.saturating_mul(diff);
// It is very unlikely that this will exist (in our poor perbill estimate) but we are giving
// it a shot.
let second_term = v_squared_2.saturating_mul(diff_squared);
if positive {
@@ -99,15 +100,15 @@ impl<T: Get<Perbill>> Convert<Fixed64, Fixed64> for TargetedFeeAdjustment<T> {
let excess = first_term.saturating_add(second_term);
multiplier.saturating_add(excess)
} else {
// Proof: first_term > second_term. Safe subtraction.
let negative = first_term - second_term;
// Defensive-only: first_term > second_term. Safe subtraction.
let negative = first_term.saturating_sub(second_term);
multiplier.saturating_sub(negative)
// despite the fact that apply_to saturates weight (final fee cannot go below 0)
// it is crucially important to stop here and don't further reduce the weight fee
// multiplier. While at -1, it means that the network is so un-congested that all
// transactions have no weight fee. We stop here and only increase if the network
// became more busy.
.max(Fixed64::from_rational(-1, 1))
.max(Fixed128::from_natural(-1))
}
}
}
@@ -119,6 +120,7 @@ mod tests {
use crate::{MaximumBlockWeight, AvailableBlockRatio, Runtime};
use crate::{constants::currency::*, TransactionPayment, TargetBlockFullness};
use frame_support::weights::Weight;
use core::num::NonZeroI128;
fn max() -> Weight {
MaximumBlockWeight::get()
@@ -129,26 +131,22 @@ mod tests {
}
// poc reference implementation.
fn fee_multiplier_update(block_weight: Weight, previous: Fixed64) -> Fixed64 {
let block_weight = block_weight as f32;
let v: f32 = 0.00004;
fn fee_multiplier_update(block_weight: Weight, previous: Fixed128) -> Fixed128 {
let block_weight = block_weight as f64;
let v: f64 = 0.00004;
// maximum tx weight
let m = max() as f32;
let m = max() as f64;
// Ideal saturation in terms of weight
let ss = target() as f32;
let ss = target() as f64;
// Current saturation in terms of weight
let s = block_weight;
let fm = v * (s/m - ss/m) + v.powi(2) * (s/m - ss/m).powi(2) / 2.0;
let addition_fm = Fixed64::from_parts((fm * 1_000_000_000_f32).round() as i64);
let addition_fm = Fixed128::from_parts((fm * Fixed128::accuracy() as f64).round() as i128);
previous.saturating_add(addition_fm)
}
fn feemul(parts: i64) -> Fixed64 {
Fixed64::from_parts(parts)
}
fn run_with_system_weight<F>(w: Weight, assertions: F) where F: Fn() -> () {
let mut t: sp_io::TestExternalities =
frame_system::GenesisConfig::default().build_storage::<Runtime>().unwrap().into();
@@ -160,7 +158,7 @@ mod tests {
#[test]
fn fee_multiplier_update_poc_works() {
let fm = Fixed64::from_rational(0, 1);
let fm = Fixed128::from_rational(0, NonZeroI128::new(1).unwrap());
let test_set = vec![
(0, fm.clone()),
(100, fm.clone()),
@@ -171,9 +169,10 @@ mod tests {
test_set.into_iter().for_each(|(w, fm)| {
run_with_system_weight(w, || {
assert_eq_error_rate!(
fee_multiplier_update(w, fm).into_inner(),
TargetedFeeAdjustment::<TargetBlockFullness>::convert(fm).into_inner(),
5,
fee_multiplier_update(w, fm),
TargetedFeeAdjustment::<TargetBlockFullness>::convert(fm),
// Error is only 1 in 10^18
Fixed128::from_parts(1),
);
})
})
@@ -184,12 +183,12 @@ mod tests {
// just a few txs per_block.
let block_weight = 0;
run_with_system_weight(block_weight, || {
let mut fm = Fixed64::default();
let mut fm = Fixed128::default();
let mut iterations: u64 = 0;
loop {
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(fm);
fm = next;
if fm == Fixed64::from_rational(-1, 1) { break; }
if fm == Fixed128::from_natural(-1) { break; }
iterations += 1;
}
println!("iteration {}, new fm = {:?}. Weight fee is now zero", iterations, fm);
@@ -217,7 +216,7 @@ mod tests {
run_with_system_weight(block_weight, || {
// initial value configured on module
let mut fm = Fixed64::default();
let mut fm = Fixed128::default();
assert_eq!(fm, TransactionPayment::next_fee_multiplier());
let mut iterations: u64 = 0;
@@ -245,79 +244,94 @@ mod tests {
#[test]
fn stateless_weight_mul() {
// This test will show that heavy blocks have a weight multiplier greater than 0
// and light blocks will have a weight multiplier less than 0.
run_with_system_weight(target() / 4, || {
// Light block. Fee is reduced a little.
// `fee_multiplier_update` is enough as it is the absolute truth value.
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::default());
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(-7500),
next,
fee_multiplier_update(target() / 4 ,Fixed128::default())
);
// Light block. Fee is reduced a little.
assert!(next < Fixed128::zero())
});
run_with_system_weight(target() / 2, || {
// a bit more. Fee is decreased less, meaning that the fee increases as the block grows.
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::default());
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(-5000),
next,
fee_multiplier_update(target() / 2 ,Fixed128::default())
);
// Light block. Fee is reduced a little.
assert!(next < Fixed128::zero())
});
run_with_system_weight(target(), || {
// ideal. Original fee. No changes.
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(0),
);
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::default());
assert_eq!(next, Fixed128::zero())
});
run_with_system_weight(target() * 2, || {
// // More than ideal. Fee is increased.
// More than ideal. Fee is increased.
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::default());
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(10000),
next,
fee_multiplier_update(target() * 2 ,Fixed128::default())
);
// Heavy block. Fee is increased a little.
assert!(next > Fixed128::zero())
});
}
#[test]
fn stateful_weight_mul_grow_to_infinity() {
run_with_system_weight(target() * 2, || {
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(10000)
);
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(10000)),
feemul(20000)
);
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(20000)),
feemul(30000)
);
// ...
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(1_000_000_000)),
feemul(1_000_000_000 + 10000)
);
let mut original = Fixed128::default();
let mut next = Fixed128::default();
(0..1_000).for_each(|_| {
next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(original);
assert_eq!(
next,
fee_multiplier_update(target() * 2, original),
);
// must always increase
assert!(next > original);
original = next;
});
});
}
#[test]
fn stateful_weight_mil_collapse_to_minus_one() {
run_with_system_weight(0, || {
let mut original = Fixed128::default(); // 0
let mut next;
// decreases
next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(original);
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default()),
feemul(-10000)
next,
fee_multiplier_update(0, original),
);
assert!(next < original);
original = next;
// keeps decreasing
next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(original);
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(-10000)),
feemul(-20000)
next,
fee_multiplier_update(0, original),
);
assert!(next < original);
// ... stops going down at -1
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(-20000)),
feemul(-30000)
);
// ...
assert_eq!(
TargetedFeeAdjustment::<TargetBlockFullness>::convert(feemul(1_000_000_000 * -1)),
feemul(-1_000_000_000)
TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::from_natural(-1)),
Fixed128::from_natural(-1)
);
})
}
@@ -326,7 +340,7 @@ mod tests {
fn weight_to_fee_should_not_overflow_on_large_weights() {
let kb = 1024 as Weight;
let mb = kb * kb;
let max_fm = Fixed64::from_natural(i64::max_value());
let max_fm = Fixed128::from_natural(i128::max_value());
// check that for all values it can compute, correctly.
vec![
@@ -339,13 +353,17 @@ mod tests {
100 * kb,
mb,
10 * mb,
2147483647,
4294967295,
MaximumBlockWeight::get() / 2,
MaximumBlockWeight::get(),
Weight::max_value() / 2,
Weight::max_value(),
].into_iter().for_each(|i| {
run_with_system_weight(i, || {
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed64::default());
let truth = fee_multiplier_update(i, Fixed64::default());
assert_eq_error_rate!(truth.into_inner(), next.into_inner(), 5);
let next = TargetedFeeAdjustment::<TargetBlockFullness>::convert(Fixed128::default());
let truth = fee_multiplier_update(i, Fixed128::default());
assert_eq_error_rate!(truth, next, Fixed128::from_parts(50_000_000));
});
});
+10 -7
View File
@@ -31,7 +31,7 @@ pub use node_primitives::{AccountId, Signature};
use node_primitives::{AccountIndex, Balance, BlockNumber, Hash, Index, Moment};
use sp_api::impl_runtime_apis;
use sp_runtime::{
Permill, Perbill, Percent, ApplyExtrinsicResult,
Permill, Perbill, Perquintill, Percent, ApplyExtrinsicResult,
impl_opaque_keys, generic, create_runtime_str,
};
use sp_runtime::curve::PiecewiseLinear;
@@ -127,8 +127,8 @@ pub const VERSION: RuntimeVersion = RuntimeVersion {
// and set impl_version to 0. If only runtime
// implementation changes and behavior does not, then leave spec_version as
// is and increment impl_version.
spec_version: 240,
impl_version: 2,
spec_version: 241,
impl_version: 0,
apis: RUNTIME_API_VERSIONS,
};
@@ -161,7 +161,8 @@ impl OnUnbalanced<NegativeImbalance> for DealWithFees {
parameter_types! {
pub const BlockHashCount: BlockNumber = 250;
pub const MaximumBlockWeight: Weight = 1_000_000_000;
/// We allow for 2 seconds of compute with a 6 second average block time.
pub const MaximumBlockWeight: Weight = 2_000_000_000_000;
pub const MaximumBlockLength: u32 = 5 * 1024 * 1024;
pub const Version: RuntimeVersion = VERSION;
pub const AvailableBlockRatio: Perbill = Perbill::from_percent(75);
@@ -180,6 +181,7 @@ impl frame_system::Trait for Runtime {
type Event = Event;
type BlockHashCount = BlockHashCount;
type MaximumBlockWeight = MaximumBlockWeight;
type DbWeight = ();
type MaximumBlockLength = MaximumBlockLength;
type AvailableBlockRatio = AvailableBlockRatio;
type Version = Version;
@@ -254,10 +256,11 @@ impl pallet_balances::Trait for Runtime {
parameter_types! {
pub const TransactionBaseFee: Balance = 1 * CENTS;
pub const TransactionByteFee: Balance = 10 * MILLICENTS;
// setting this to zero will disable the weight fee.
pub const WeightFeeCoefficient: Balance = 1_000;
// In the Substrate node, a weight of 10_000_000 (smallest non-zero weight)
// is mapped to 10_000_000 units of fees, hence:
pub const WeightFeeCoefficient: Balance = 1;
// for a sane configuration, this should always be less than `AvailableBlockRatio`.
pub const TargetBlockFullness: Perbill = Perbill::from_percent(25);
pub const TargetBlockFullness: Perquintill = Perquintill::from_percent(25);
}
impl pallet_transaction_payment::Trait for Runtime {