SIBERIA BENCH

blaster · observer · coordinator
— TPS h —
Finalized TPS
—
Transactions / second
Block interval
—
Average production interval
Commit latency P95
—
Proposal to commit
Finality latency P95
—
Transaction to finality
idle no active run — start one on Configure

Submitted vs committed vs finalized TPS (per second)

Block production interval (ms between committed blocks)

Mempool backlog (transactions waiting, deepest node)

Commit latency (proposal → commit, ms)

Transaction finality latency (submitted → final, ms)

Cluster

1Configure a run
a preset, or your own — the estimate checks it first
2Watch it live
lines overlapping = chain keeping up
3Read the headline
Finalized TPS · finality · loss · divergences

Validators

NodeRPCWSHeightFinalizedSlotLagCPURSS

Finalized TPS (10 s rolling, unique tx hashes — hover to inspect)

Block production

Alerts

  • no alerts

Event log (live WS events, newest last)

Preset

Estimate (arithmetic only — nothing is sent until Start)

pick a preset or edit the form — the verdict appears here.

Start jumps to the Monitor. You can stop & drain at any time — already-finalized transactions are never lost.

How TPS and finality are calculated

Offered — transactions the scheduler attempts to hand to workers. Submission — requests that actually left the generator. Accepted — sendTransaction acknowledged (mempool admission). Included — unique tx hashes observed in proposed blocks. Executed-OK — equal to Included in Siberia (blocks execute atomically). Finalized — unique tx hashes in 3-chain-final blocks.

Every count is keyed by unique tx hash (multi-node observation never double counts). Steady state = end of warm-up → end of load. Drain is reported separately. Siberia finality is deterministic BFT 3-chain: a block is final when it, its parent and its grandparent are QC-certified (quorum 3-of-4).

One in-flight transaction per sender wallet — this blaster's own closed loop. W1 replaced the chain's sequential-nonce rule with a recent-blockhash replay window (many in-flight txs per sender are legal on-chain now), but the generator kept its pre-W1 discipline while NONCE_PLAN ticket F widens it. The cycle is ~1.5 slots (1.5 s at 1 s slots, 0.75 s at the 500 ms compose slot), so you need roughly target × 1.5 × slot-seconds senders for a target TPS; the estimate suggests the exact count when you fall short.

Nonce-wait counters are that closed loop's pacing, never errors.

Derived sender accounts

Senders are derived deterministically from (seed, index) — the same seed always reproduces the same accounts, so balances carry across runs. Balances are read live from the cluster.

Unfunded accounts are normal for amount = 0 runs — zero-value transfers need no balance. Fund only when a run uses real amounts.

How funding works

Funding drips requestFaucetFunds per account — the exact same path the run's funding phase uses (retried with backoff, concurrency bounded to be gentle). Pre-funding here makes a later run's funding phase a no-op top-up.

The faucet drips one request at a time (the same one-in-flight closed loop), so funding a large population is inherently serial and takes roughly one cycle (~1.5 slots; ~0.75 s at the 500 ms compose slot) per account. For throughput measurements prefer amount 0 runs, which skip funding entirely.

Amounts are additive: funding an account twice adds twice. Balances update as faucet transactions finalize (a few seconds after the drip is accepted).

Accounts (live from the cluster)

#AddressBalance
loading…

Run

Throughput (per-second deltas, unique hashes — click legend to toggle · drag to zoom · double-click to reset)

Latency percentiles (ms, submit → stage — hover a bar)

Pending & nonce waits

What to watch

Throughput chart: when all five lines overlap, the chain is keeping up with everything the generator offers — the run measures the chain at your target rate, not its maximum. A persistent gap between offered (grey) and finalized (green) means the chain, the host, or the sender population is the bottleneck. Pending climbing = in-flight backlog growing. Nonce waits climbing = the schedule is ahead of the wallets' K-in-flight gate (add senders, raise K, or lower the target). Rejected / failed anything above zero deserves a look at the validators.

Runs (live + archived from disk — click a row for its headline, check two or more to compare)

RunNameStatusSubmittedFinalizedFinalized TPSUnresolvedCreated
pick two or more runs — metrics table + overlaid throughput curves

Headline (the numbers to quote)

select a run above

Latency drill-down (click a run to load its stage ladder)

select a run above

Run detail

select a run above