Benchmark reports

The Performance & Tracing team benchmarks the Leios node on the same cluster that benchmarks every cardano-node release. The builds here are release candidates of the Leios node, ahead of its integration into a cardano-node release. Each run below holds the endorser-block parameters fixed and saturates them with value transactions, so a change between runs is attributable to the parameter or the build that changed.

These reports will move to Cardano Updates, where the team publishes its release benchmarks. Until then, they are published here.

Endorser-block reports

Each report follows the endorser blocks of one run through five stages (forging, propagation, acquiring, closures, certifying) and closes with the forks of the ranking-block chain.

Reading the certification rate

The EB certification rate is the share of announced endorser blocks that were certified. The expected certification rate is the share the protocol's timing allows on average: an endorser block can be certified only if no other ranking block arrives within its 14-slot window, which happens for (1 − 0.05)¹⁴ = 48.8% of them. 🟢 marks a run within 5 points of the expected rate, 🟡 within 15 points, 🔴 further below.

w45 and the voting fix

Two runs of the 100 kB / 1 MB parameter set with 374 B transactions (two inputs, two outputs) on 51 nodes: the Leios node build w45 as released, and w45 with the voting fix that keeps the input table of an endorser block's transactions outside the ledger state (ouroboros-consensus #2498). With the fix, 49.0% of announced endorser blocks were certified, up from 26.0%, and the voters that declined an endorser block as too late fell from 56,472 to 104.

Build	Parameter set (reference cap / closure cap)	Transactions	Run	EB certification rate / expected certification rate	Report
w45 with the voting fix	100 kB / 1 MB	374 B, 2 inputs 2 outputs	2026-11-09	🟢 49.0% / 48.8%	open
w45	100 kB / 1 MB	374 B, 2 inputs 2 outputs	2026-11-06	🔴 26.0% / 48.8%	open

In the run with the voting fix, one node disconnected a couple of hours before the end, so some maximum columns are off.

w42

All four runs used the Leios node build w42 and value transactions, in two shapes. Transactions of 200 B, with one input and one output, fill the reference cap first; transactions of 374 B, with two inputs and two outputs, the workload of the team's Praos comparison runs, fill the closure cap first. Each parameter set therefore has a run at each of its two limits, and each report names the other run of its pair as its control.

Build	Parameter set (reference cap / closure cap)	Transactions	Run	EB certification rate / expected certification rate	Report
w42	40 kB / 260 kB	200 B, 1 input 1 output	2026-10-22	🟢 49.8% / 48.8%	open
w42	40 kB / 260 kB	374 B, 2 inputs 2 outputs	2026-10-23	🟢 51.8% / 48.8%	open
w42	100 kB / 1 MB	200 B, 1 input 1 output	2026-10-20	🟡 36.6% / 48.8%	open
w42	100 kB / 1 MB	374 B, 2 inputs 2 outputs	2026-10-21	🔴 27.4% / 48.8%	open