Electra: performance issue computing sync committee indices

[twoeths]

Describe the bug

as informed by @dapplion it could be much slow to compute sync committee indices in electra, theorically 64x slower

lodestar/packages/state-transition/src/util/seed.ts

Line 105 in 0ac8bb9

const MAX_EFFECTIVE_BALANCE_INCREMENT =

Expected behavior

• need to write a benchmark for it when all validators have effectiveBalanceIncrement = 32 (worse case scenario)
• improve it. 2 ways i can think of:
  • use hashtree. Considering the below for loop, we can compute randByte ahead for the next n loops

 while (syncCommitteeIndices.length < SYNC_COMMITTEE_SIZE) {
    const shuffledIndex = computeShuffledIndex(i % activeValidatorCount, activeValidatorCount, seed);
    const candidateIndex = activeValidatorIndices[shuffledIndex];
    const randByte = digest(
      Buffer.concat([
        seed,
        //
        intToBytes(Math.floor(i / 32), 8, "le"),
      ])
    )[i % 32];

    const effectiveBalanceIncrement = effectiveBalanceIncrements[candidateIndex];
    if (effectiveBalanceIncrement * MAX_RANDOM_BYTE >= MAX_EFFECTIVE_BALANCE_INCREMENT * randByte) {
      syncCommitteeIndices.push(candidateIndex);
    }

    i++;
  }

- otherwise go native

Steps to reproduce

No response

Additional context

No response

Operating system

Linux

Lodestar version or commit hash

v1.25.0

[twoeths]

got 10s processSyncCommitteeUpdates() in peerdas-devnet-4, seems like due to electra

[twoeths]

also found this in pectra-devnet-6