Summary
Drive Heima's parachain block time from ~6s to ~2s using elastic scaling on 3 coretime cores, giving latency-sensitive EVM applications Base-class responsiveness while keeping Heima's much cheaper gas.
Why
Per-op gas on Heima is already excellent — the historic heavy cost (the pure-Solidity P-256 passkey verify, ~1M gas) is gone now that RIP-7212 is live (runtime 9261): K11 verify is ~28k gas. The only axis where an L2 like Base still wins is latency: Base's sequencer produces ~2s blocks vs Heima's ~6s. For interactive flows a user-facing write costs ≈ one block of wait, so block time is the perceived latency.
Current state (please confirm)
Heima appears to run ~6s blocks, which suggests async backing is already enabled (the 12s→6s win is banked). Confirming this is step 1.
Mechanism
Heima is a Polkadot parachain (para_id 2013). The relay already ships async backing + elastic scaling + agile coretime (Polkadot SDK 2509, Oct 2025), so the capability is present — the remaining work is Heima-side (runtime + collator) plus acquiring cores.
A parachain block is capped at 2s of execution; the relay slot is 6s. Elastic scaling lets one parachain occupy multiple cores in parallel, one block per core per slot:
| Cores |
Blocks / 6s relay slot |
Block time |
| 1 (today) |
1 |
~6s |
| 2 |
2 |
~3s |
| 3 |
3 |
~2s |
This also raises aggregate blockspace ~3× — throughput and latency both improve, it is not a trade.
Proposed change (runtime + collator)
| Knob |
Today |
Target |
MILLISECS_PER_BLOCK / SLOT_DURATION |
6000 |
2000 |
BLOCK_PROCESSING_VELOCITY |
1 |
3 |
UNINCLUDED_SEGMENT_CAPACITY |
~2–3 |
raise to match velocity (≈ velocity · relay/para + 1) |
RELAY_CHAIN_SLOT_DURATION_MILLIS |
6000 |
6000 (fixed by relay) |
| Collator binary |
lookahead collator |
slot-based collator (required for >1 core) |
| Coretime (Agile Coretime) |
1 core |
3 cores on para 2013 |
The permissioned (Litentry-run) collator set qualifies for the elastic-scaling MVP (trusted-collator phase — no candidate-receipt change). Validate on Heima Paseo (para_id 2013 testnet) before mainnet.
Diagnostics requested
- Current mainnet
MILLISECS_PER_BLOCK.
- Polkadot SDK / runtime version (slot-based collator needs ≥ the 2509 line).
- Collator type (lookahead vs slot-based).
- Cores currently assigned to para 2013.
- Budget appetite for 3 cores.
Costs / risks
- ~3× recurring coretime spend (3 cores on Polkadot).
- Collator hardware: 2s blocks at ~100% duty cycle need collators that import + author a 2s-weight block in < 2s wall-clock.
- Coordinated runtime + collator upgrade, tested on Paseo first.
References
Summary
Drive Heima's parachain block time from ~6s to ~2s using elastic scaling on 3 coretime cores, giving latency-sensitive EVM applications Base-class responsiveness while keeping Heima's much cheaper gas.
Why
Per-op gas on Heima is already excellent — the historic heavy cost (the pure-Solidity P-256 passkey verify, ~1M gas) is gone now that RIP-7212 is live (runtime 9261): K11 verify is ~28k gas. The only axis where an L2 like Base still wins is latency: Base's sequencer produces ~2s blocks vs Heima's ~6s. For interactive flows a user-facing write costs ≈ one block of wait, so block time is the perceived latency.
Current state (please confirm)
Heima appears to run ~6s blocks, which suggests async backing is already enabled (the 12s→6s win is banked). Confirming this is step 1.
Mechanism
Heima is a Polkadot parachain (para_id 2013). The relay already ships async backing + elastic scaling + agile coretime (Polkadot SDK 2509, Oct 2025), so the capability is present — the remaining work is Heima-side (runtime + collator) plus acquiring cores.
A parachain block is capped at 2s of execution; the relay slot is 6s. Elastic scaling lets one parachain occupy multiple cores in parallel, one block per core per slot:
This also raises aggregate blockspace ~3× — throughput and latency both improve, it is not a trade.
Proposed change (runtime + collator)
MILLISECS_PER_BLOCK/SLOT_DURATIONBLOCK_PROCESSING_VELOCITYUNINCLUDED_SEGMENT_CAPACITYvelocity · relay/para + 1)RELAY_CHAIN_SLOT_DURATION_MILLISThe permissioned (Litentry-run) collator set qualifies for the elastic-scaling MVP (trusted-collator phase — no candidate-receipt change). Validate on Heima Paseo (para_id 2013 testnet) before mainnet.
Diagnostics requested
MILLISECS_PER_BLOCK.Costs / risks
References
docs/research/heima-block-speed.md.