Stage 1 of the latency programme: five header-delay arms on a ten-region fleet

2026-10-06 · started 20:59 UTC · ended 2026-10-07 00:52 UTC · merged mixed window 9 min 35 s

telcoin-network ba43a1f0e · devnet-bench-10-5-2026 · commit verified from node startup logs

78 node starts across 11 instances logged telcoin-network 0.1.0 (ba43a1f0e270ca1c572d1b8378b39d2fc6360b57) The node image probe recorded the same commit before deploy Node logs: fork schedule: seed_signature, multi_workers, prevrandao, leader_seeded_ordering, and subsecond_timestamp active from genesis Commit ba43a1f0e is the head of the public integration branch devnet-bench-10-5-2026 after the merge of fix/certifier-keep-inflight-on-repropose into the 6 October build 964087e02: the certifier keeps its in-flight vote collection when the proposer re-sends an identical header. It is the build of r20261006-0458, unchanged.

We moved the benchmark fleet from five regions with two validators each to ten regions with one validator each (Los Angeles, Montréal, London, the Netherlands, Mumbai, Singapore, Hong Kong, Tokyo, Seoul and Sydney), kept the certifier-fix build of the 6 October re-run (r20261006-0458) and ran five consensus-timing arms in one deployment with chaos off. Each arm set the minimum and maximum header delay on every node live, by a rolling restart, and was followed by about eight minutes of idle chain and one blast of 480 s of mixed load: 250/500 ms (A0, the timing of the earlier runs), 50/250 (A1), 400/800 (A2), 250/1000 (A3) and 500/1000 (A4). This page publishes the A3 blast; the arms table under Consensus lists all five. The idle round was 527 to 550 ms in every arm: the minimum header delay does not pace the idle round, which is set by the arrival of the seventh certificate about 525 ms after a proposal, and raising the maximum header delay from 500 to 1,000 ms removed the re-proposal churn (92 re-proposals a minute per node in A0, about 5 in A2 to A4) without changing the round. Under load every arm was execution-bound at 1.22 to 1.41 rounds/s and 14.4 to 15.8 blocks/s with a median commit latency of 1.65 to 1.75 s; A3 had the fastest loaded cadence and the fewest re-proposals. No arm met the programme's criterion that every validator certify at least 90 % of its own headers while idle: London certified 7 to 50 % and the Netherlands 0.7 to 21 % in every arm, because a vote request took about twice the network round trip on every link, and longer still from Europe to Asia, so their sixth vote arrived after the next round's parents had and they cancelled their own header. That cost is a protocol-negotiation round trip paid by every vote request, not a timer value; the next run carries a node image that removes it.

Compared against r20261006-0458.

Key figures

5,831.1tx/s
Average chain throughput, all load generators running
all generators active
6,623.1tx/s
Best 60 seconds, all load generators running
all generators active
7.95s
Time to land, p95
whole run
2.46–4.92s
Commit latency, p95
merged mixed window

How it was run

Software

Upstream committelcoin-network ba43a1f0e270ca1c572d1b8378b39d2fc6360b57 · devnet-bench-10-5-2026 · commit verified from node startup logs
Node imageus-docker.pkg.dev/telcoin-network/tn-public/adiri:mmap-subsecond-certfix-10-6-2026
Node image digestsha256:caeb6be2f2732658841fa760b75e1adcec37d3a966a413fb48be8d1b9c32313e
Node version0.1.0
Fork schedulesubsecond-from-genesis
Load generator commit9f61303
Load generator image digestsha256:515311a6a8ffdde56ed190f85c11cee6abdec1434e2fff5e8ffd57d4a554eee2
Analysis commit9f61303
Harness commitaccd5d9abb41fab394d88ab4ef73c38eb74f0e49
Harness submodulescommon: b498b71e107f09c7e6a7a68ae5ba71a514b98609
tn-node-deployment: 1f0385d79a99e008cadf66f0522f4417a09e7543

Fleet

Fleetc3-highcpu-8 + Hyperdisk Balanced (c3-hdb), one validator per region
Validators10
Observers1
Machine typec3-highcpu-8
vCPUs per node8
Memory per node16 GB
Diskhyperdisk-balanced, 100 GB, 20,000 IOPS, 800 MiB/s
Regionsasia-east2 (Hong Kong), asia-northeast1 (Tokyo), asia-northeast3 (Seoul), asia-south1 (Mumbai), asia-southeast1 (Singapore), australia-southeast1 (Sydney), europe-west2 (London), europe-west4 (Netherlands), northamerica-northeast1 (Montréal), us-west2 (Los Angeles) (10)
Load generators10 generators × e2-standard-2 (the Sydney generator n2d-standard-2) × 600 senders each
Load placementone blaster per validator, in that validator's zone, sending only to it

Chain

Chain ID32286
Networkbench
Epoch length1,200 s (20 min)

Consensus parameters

Max header delay500 ms
Min header delay250 ms
Max batch delay250 ms
Vote timeout5,000 ms
Parameters fileconsensus-parameters.yaml
Confirmed bynode startup logs

Parameter arms

ArmFrom (UTC)To (UTC)Values
a02026-10-06 21:58:402026-10-06 22:38:01all 11 nodes: max_header_delay=500ms, min_header_delay=250ms
a12026-10-06 22:38:212026-10-06 23:18:45all 11 nodes: max_header_delay=250ms, min_header_delay=50ms
a22026-10-06 23:19:052026-10-06 23:58:56all 11 nodes: max_header_delay=800ms, min_header_delay=400ms
a32026-10-06 23:59:162026-10-07 00:39:10all 11 nodes: max_header_delay=1000ms, min_header_delay=250ms
a42026-10-07 00:39:30—all 11 nodes: max_header_delay=1000ms, min_header_delay=500ms

Workload

Warm-up60 s (1 min)
Mixed phase480 s (8 min)
Cool-down60 s (1 min)
Transaction typesbatched_self_send, contract_deploy, faucet_drip, native_transfer, stablecoin_approve, stablecoin_transfer, uniswap_v2_add_liquidity, uniswap_v2_remove_liquidity, uniswap_v2_swap, uniswap_v3_swap, uniswap_v4_swap, wtel_unwrap, wtel_wrap

Steps

  1. Before creating anything, the harness resolved the node image tag adiri:mmap-subsecond-certfix-10-6-2026 to its registry digest and pinned the run to that digest (sha256:caeb6be2…), the image of r20261006-0458. It ran the pinned image to confirm that it reports telcoin-network 0.1.0 built from commit ba43a1f0e, has the sub-second timestamp fork active from genesis and contains the log messages later checks rely on, and it recorded the committed consensus parameters file and its three delays.
  2. Preflight checked access, tools, quotas and a clean genesis directory, and confirmed the load-generator image built from tn-transaction-generator commit 9f61303, which repairs gaps in the senders' setup transactions and hands the unlanded steps back to the senders (the 6 October runs used 3644adc).
  3. The harness created ten validators and one observer on c3-highcpu-8 (8 vCPU, 16 GB), each with a 100 GB Hyperdisk Balanced boot disk provisioned at 20,000 IOPS and 800 MiB/s, one validator in each of ten regions (Los Angeles, Montréal, London, the Netherlands, Mumbai, Singapore, Hong Kong, Tokyo, Seoul and Sydney) with the observer in Los Angeles, and opened the consensus ports between the nodes only. The nodes reach each other over their public addresses.
  4. It generated validator keys and built genesis for a separate benchmark chain (chain id 32286, 20-minute epochs).
  5. It finalized each node's configuration with the committed benchmark consensus parameters: maximum header delay 500 ms, minimum header delay 250 ms and maximum batch delay 250 ms (the A0 values), with every other key as in the devnet defaults. The file sets no vote timeout, so the node's built-in 5 s applied. The node log filter was raised to debug for the certifier and proposer modules, info elsewhere, so that every vote request and its outcome is logged. It then started the chain.
  6. It relabelled each node's metrics agent so that every sample from these nodes is recorded under the benchmark network, never the live devnet.
  7. It restarted the nodes one at a time with RPC open on the internal network, waiting after each until the node answered with the benchmark chain id and consensus rounds were advancing. Each node's boot lines had to show the pinned commit, the three consensus delays, the 5,000 ms vote timeout and the fork schedule recorded at the gate, and all eleven did. An on-node sampler was then started on each node. The fleet was healthy at 21:40 UTC.
  8. A health gate confirmed that all eleven nodes reported to Prometheus under the benchmark network with consensus rounds rising, and a read of the ConsensusRegistry through an RPC tunnel confirmed the 1,200 s epoch duration.
  9. It deployed the devnet contract set that the mixed workload calls.
  10. The five arms then ran in order from 21:58 to 01:18 UTC: A0 250/500 ms (21:58 to 22:38), A1 50/250 (22:38 to 23:18), A2 400/800 (23:19 to 23:58), A3 250/1000 (23:59 to 00:39) and A4 500/1000 (00:39 to 01:17). For each arm after A0 the harness rewrote the two header delays on every node with a fleet-wide apply, restarting the eleven nodes one at a time (16 to 20 s each, about three minutes for the fleet) and waiting for each to log the new values and rejoin consensus, and recorded the arm's begin, end and each node's values in the manifest. A0 was the deployed configuration, so no node was restarted for it and its record was entered by hand in the same form.
  11. Each arm had about eight minutes of idle chain after its apply, measured from the last node's confirmation to the start of the funding step. A one-off funder VM then funded 600 sender accounts for each of ten load generators, one per validator in that validator's zone and sending only to it, and the generators started together. Nine ran on e2-standard-2; the Sydney generator fell back to n2d-standard-2 on every blast because e2 and n2 machines were out of stock in its zone. Each generator ran 60 s of warm-up, 480 s of mixed load across 13 transaction types and 60 s of cool-down. No chaos action was scheduled and no network delay was injected.
  12. The blast published here is the fourth, in arm A3: generators started at 00:17:04 UTC, mixed phase 00:19:34 to 00:27:34. After each blast the harness collected each generator's results and merged them with aggregate.py, derived the sizing figures with derive-sizing.py, and pulled 60-second machine metrics from Cloud Monitoring and 15-second node metrics from Prometheus; the records of each finished blast were kept apart when the next began.
  13. At the end of the run the harness recorded, for every validator, the share of its own headers that reached a quorum in each arm and each phase of the last blast, and pulled a log snapshot from each validator that fell under 90 % in any window (London, the Netherlands, Mumbai and Sydney).
  14. Teardown began at 01:27 UTC after a local gate had verified the saved results (767 files, 1.6 GB). It saved every node's logs, sampler files and service journal, then deleted the fleet; the final index of the results was written at 01:40 UTC. The publishing pipeline then read each node's logged commit, consensus delays, vote timeout and fork schedule back from those logs.

The harness scripts live in a private repository; the steps above describe what they do.

Results

MetricValueWindowSourceMethod
Average chain throughput, all load generators running5,831.1 tx/sall generators activecombined.series.per_second[chain_tx]Chain transactions per block-timestamp second, summed over the 385-s window when every load generator was in its mixed phase and divided by its length.
Best 60 seconds, all load generators running6,623.1 tx/sall generators activecombined.series.per_second[chain_tx]Highest mean chain tx/s over any 60 consecutive seconds inside the all-generators window (aggregate.py best_window_mean).
Transactions landed over the whole blast3,567,048 txwhole runcombined.totals.landed_chainSum of tx_count over every block the load generators saw during the blast, own and foreign transactions alike.
Blocks per second15.62 blocks/smerged mixed windowcombined.blocks.rows[timestamp]Chain blocks whose timestamp falls in the merged mixed window, divided by its length (aggregate.py landed.capacity.blocks_per_sec over the same window).
Time to land, median3.17 swhole runcombined.latency.ts_ms.p50Block timestamp minus submit time over every landed own transaction in the blast, from the load generators' merged latency histograms (aggregate.py merge_latency), in seconds.
Time to land, p957.95 swhole runcombined.latency.ts_ms.p95Block timestamp minus submit time over every landed own transaction in the blast, from the load generators' merged latency histograms (aggregate.py merge_latency), in seconds.
Commit latency, p952.46–4.92 smerged mixed windowmetrics.prometheus.series.tn_primary_commit_latency_seconds.by_node[healthy validators].p95For each healthy validator, the nearest-rank p95 over the merged mixed window of its 15-s histogram_quantile(0.95) commit-latency series; reported as the range over validators.
Execution time per consensus output1.39–1.44 smerged mixed windowmetrics.prometheus.series.tn_engine_execution_duration_seconds.by_node[healthy validators].meanFor each healthy validator, the nearest-rank median over the merged mixed window of its 15-s mean execution-duration series; reported as the range over validators.
Write throughput, p95, busiest validator316.8 MiB/smerged mixed windowmetrics.gcp[bench-validator-07].disk_write_bytes_per_sec p95Nearest-rank p95 of the 60-s mean write-byte rate over the merged mixed window for each healthy validator; the highest (bench-validator-07) is reported.
Write operations, p95, busiest validator20,105 IOPSmerged mixed windowmetrics.gcp[bench-validator-04].disk_write_ops_per_sec p95Nearest-rank p95 of the 60-s mean write-operation rate over the merged mixed window for each healthy validator; the highest (bench-validator-04) is reported.
CPU, p95, busiest validator35 %merged mixed windowmetrics.gcp[bench-validator-06].cpu_utilization p95Nearest-rank p95 of the 60-s CPU utilisation over the merged mixed window for each healthy validator; the highest (bench-validator-06) is reported.
Node resident memory, p95, busiest validator6.35 GiBmerged mixed windowmetrics.gcp[bench-validator-10].telcoin_rss_bytes p95Nearest-rank p95 of the 60-s telcoin process resident memory over the merged mixed window for each healthy validator; the highest (bench-validator-10) is reported.
IO pressure, highest reading, range over validators11.9–13.8 %merged mixed windowsampler[healthy validators].psi_io_some_avg10 maxFor each healthy validator, the highest 10-s 'some' IO pressure reading the on-node sampler logged inside the merged mixed window; reported as the range over validators.

Chain TPS per 5-minute interval

Chain TPS per 5-minute interval 2 intervals from 00:18 to 00:27 UTC; average 4,933 TPS, highest 5,736 TPS. The merged mixed window ran from 00:18 to 00:27 UTC. TPS 0 1,000 2,000 3,000 4,000 5,000 6,000 00:18–00:23 UTC: 5,736 TPS, 9.1 generators active 00:23–00:27 UTC: 4,129 TPS, 7.4 generators active avg 4,933 00:18 UTC 00:23 UTC

Each bar is the average chain TPS over one 5-minute interval of the merged mixed window (00:18–00:27 UTC); fainter bars had fewer generators active than the maximum and the dashed line is the average of all bars.

5-minute data (2 rows)
start_utcend_utcgenerators_activechain_tps
2026-10-07T00:18:19Z2026-10-07T00:23:19Z9.15,736.5
2026-10-07T00:23:19Z2026-10-07T00:27:54Z7.44,128.8

The fine-grained chain throughput series (its step_s field gives the seconds per point) is published as series/tps-1s.json.

Other networks

MeasureTelcoin Network (this run)Tempo
Throughput5,831.1 tx/s (Avg chain TPS)16,455 TPS, shown as "Settled TPS" for the nightly run of 5 Oct 2026 (commit de5f5d3). perf.tempo.xyz describes it as "Median settled throughput across 130 measured blocks."
How throughput is measuredChain transactions per block-timestamp second, summed over the 385-s window when every load generator was in its mixed phase and divided by its length.The methodology page defines TPS as "Transactions per second (derived from gas / avg tx gas)": gas used divided by the average gas of one transaction, reported as the median over the measured blocks.
Run length9.6 min merged mixed window; each generator was set to a 8 min mixed phase2m 39s for the 5 Oct 2026 nightly run (130 measured blocks at a 687 ms average block time), per perf.tempo.xyz.
Validators and placement10 validators and 1 observer across 10 regions: asia-east2 (Hong Kong), asia-northeast1 (Tokyo), asia-northeast3 (Seoul), asia-south1 (Mumbai), asia-southeast1 (Singapore), australia-southeast1 (Sydney), europe-west2 (London), europe-west4 (Netherlands), northamerica-northeast1 (Montréal), us-west2 (Los Angeles); load generators: one blaster per validator, in that validator's zone, sending only to itperf.tempo.xyz labels the scenario "Public mix · multi-region". The methodology page says "Benchmarks run two local validators in consensus on a dedicated bare-metal server with 100 GiB pre-bloated state." The public tempoxyz/tempo repo has nightly workflows for both kinds of setup: bench-e2e runs on one bare-metal runner, and bench-e2e-multi-region defaults to 10 validators on cloud VMs (its regions input reads "Empty uses the selected cloud's five defaults."). The run page does not say which setup produced this number.
Hardwarec3-highcpu-8, 8 vCPUs, 16 GB memory; disk: hyperdisk-balanced, 100 GB, 20,000 IOPS, 800 MiB/sThe methodology page lists a dedicated bare-metal server: "AMD EPYC 4585PX (16C/32T, 4.3/5.7 GHz)", "128 GB 3600 MHz" memory and "2×960 GB NVMe + 2×1.92 TB NVMe via schelk". The repo's multi-region workflow uses cloud VMs instead (c4d-standard-48-lssd on GCP, c6id.8xlarge on AWS).
Workloadbatched_self_send, contract_deploy, faucet_drip, native_transfer, stablecoin_approve, stablecoin_transfer, uniswap_v2_add_liquidity, uniswap_v2_remove_liquidity, uniswap_v2_swap, uniswap_v3_swap, uniswap_v4_swap, wtel_unwrap, wtel_wrap; 10 generators × 600 senders eachperf.tempo.xyz labels the scenario "Public mix"; the methodology page says "100% TIP-20 token transfers (ERC-20 equivalent)". The run used 1,000 accounts and 4 TIP-20 tokens.
StateFresh chain from genesis for every runThe methodology page says "100 GiB pre-bloated state".
Latency (as each source defines it)Time to land, median: 3.17 s; Time to land, p95: 7.95 s; Commit latency, p95: 2.46–4.92 sBlock time of 687 ms, which the methodology page defines as "Average wall-clock time between blocks". The run page gives no per-transaction latency. Separately, tempo.xyz says "Blocks finalize in ~0.6 seconds with no re-orgs." (a network claim, not a figure from this benchmark).
Source and fetch dateThis page; every figure names its source field.perf.tempo.xyz nightly run of 5 Oct 2026 (commit de5f5d3), the perf.tempo.xyz methodology page, tempo.xyz, and the benchmark workflows in the tempoxyz/tempo repo. All fetched on 5 Oct 2026.

Sources for Tempo (fetched 2026-10-05):

Tempo's setup may not match ours: its methodology page describes two validators on one bare-metal server, while our runs use ten validators across five regions, so read this table as context, not a ranking.

This table is context, not a ranking.

Caveats

  • No arm met the pass criterion of the programme, that every validator certify at least 90 % of its own headers while the chain is idle. London (bench-validator-02) certified 7 to 50 % of its own headers idle and the Netherlands (bench-validator-06) 0.7 to 21 %, in every arm; in A3, 15 % and 7 %. They held 0 to 2 % of the leader slots against a fair share of 10 %, and 75 to 100 % of the certificates they did form were never committed, because they formed after the round had moved on. The other eight validators certified 91.5 to 100 % idle (Sydney 92.5 % in A0 and 99.7 to 99.9 % in the others; Mumbai 91.5 % in A0 and at least 98.8 % elsewhere). Under load London and the Netherlands reached 59 to 85 % and Mumbai 74 to 90 %, never 90 %.
  • The cause is in the vote path, not in a timer value, and it was the same in all five arms. The time from a vote request to its answer is fixed per link (median and 90th percentile within 1 to 3 ms) and symmetric. From Los Angeles the nine peers answer in 153 to 450 ms and the sixth vote is in at 291 ms; from London the sixth vote (Sydney's) is in at 554 ms and from the Netherlands at 577 ms, because the Asian validators answer those two in 472 to 724 ms (Tokyo 472, Seoul 523, Sydney 554, Hong Kong 559, Singapore 594, Mumbai 724) against 204 to 450 ms when Los Angeles asks them. A node proposes the next round as soon as it holds seven certificates for the current one; London and the Netherlands receive the seventh other certificate 526 to 552 ms after their own proposal, before their own sixth vote, propose, and cancel their own header. The cut-off is sharp: own headers alive for under 540 ms certified 0 %, over 575 ms 90 to 100 %, and every cancelled Netherlands header in the A3 idle window had three to five votes, never six (841 of 841). Two causes were then found in the node source and confirmed against the run's own per-link timings. First, every vote request opens a new libp2p substream and negotiates its protocol before sending the request, so each vote costs an extra network round trip: from Los Angeles all nine links measure 2.01 to 2.08 times the probed UDP round trip, while certificates travel on gossip's long-lived streams and arrive in half a round trip, which is why parents beat votes. Second, on the Europe-to-Asia links there is a further wait on the voter's side (Hong Kong, Singapore and Mumbai answer London 155, 256 and 416 ms after the request arrives while answering Los Angeles at once) whose mechanism this run could not show, because the vote-request handler was not in the debug filter. A detour over the nodes' public addresses was ruled out on 7 October by a probe of every pair on public and private addresses in one run: equal on all 90 ordered pairs (median 0.6 ms apart, none more than 15 ms). The next run's node image removes the negotiation round trip.
  • The idle round was 527 to 550 ms in every arm (A0 550, A1 538, A2 527, A3 530, A4 527 ms median; 555 to 756 ms at the 95th percentile). A minimum header delay of 50 ms gave 538 ms and 250 ms gave 530 to 550 ms, as the model predicted; 400 and 500 ms did not raise the round either, which the model had not predicted: the minimum timer never bound, because the seventh certificate arrives about 525 ms after a proposal. With London and the Netherlands out of the quorum pool the seventh-fastest certificate is Mumbai's, at about 455 ms, which sets the round at about 530 ms; the model had predicted about 400 ms with all ten validators certifying. The maximum header delay changed only the churn: at 500 ms the timer fired every round (92 re-proposals a minute per node in A0, 194 in A1 at 250 ms); at 1,000 ms about 5 (A2 5.7, A3 4.9, A4 5.1 idle; 23, 9.4 and 13.8 under load).
  • Each node's time from its own proposal to its own quorum (median, idle) was the same in every arm and 20 to 170 ms slower than the model's fitted constants: Seoul 264 ms, Tokyo 272, Los Angeles 297, Hong Kong 297, Singapore 344, Montréal 406, Sydney 427, Mumbai 449 to 467, London 556 to 564, the Netherlands 580 to 585. Under load the medians rose 10 to 80 ms for the near nodes and the 95th percentiles doubled. Mumbai's zone was not in the latency probe, and its value is 160 ms above the model; Mumbai also had 11 to 21 % of its certificates never committed in A2 to A4, the most after London and the Netherlands.
  • Under load every arm was execution-bound and the arms barely differ: 1.22 to 1.41 rounds/s, 14.4 to 15.8 blocks/s, median commit latency 1.65 to 1.75 s and 95th percentile 2.42 s in every arm, by block-rate over the generators' block lists. By that method A3 was the fastest arm, 1.41 rounds/s, 15.84 blocks/s and 5,005 tx/s (A0 4,806, A1 4,238, A2 4,839, A4 4,271), and the only arm whose loaded time-to-own-quorum p95 for Sydney (593 ms) stayed under the median round (630 ms). The KPIs on this page use the pipeline's own windows and methods, so their values differ from the block-rate figures. The generators landed 98.1 to 98.5 % of what they sent in every arm.
  • The offered load was uneven for a consensus reason. The generators pinned to London, the Netherlands and Mumbai submitted 20,000 to 160,000 transactions per arm against 280,000 to 740,000 for the other seven, and landed 51 to 95 % of their own. Their batches sit in headers that are cancelled, so the validator's pool drops the head of the generator's queue and the generator re-sends it (in the published blast the Netherlands generator submitted 111,689 transactions, 101,863 of them landed and 9,826 never did). The three validators with the lowest certification are exactly the three generators with the lowest volume, and the all-generators throughput on this page is lower than the fleet could carry for that reason.
  • London's commit latency 95th percentile of 4.04 s in A4 (median 1.88 s) came from a late epoch close, not from the timers: from 01:04:09 to 01:07:08 UTC it proposed with 2 to 3 s gaps and 114 of its vote requests were rejected because it was still proposing in epoch 10 when its peers expected epoch 11; it rejoined at the first round of epoch 11 after a 16 s gap. No other node had any vote error other than the cancellations, in any window, and there was no commit stall in any arm.
  • The consensus values at the top of this page (maximum header delay 500 ms, minimum 250 ms) are those of the committed parameters file, which the nodes started with and ran in A0. The blast published here ran in A3, with a maximum header delay of 1,000 ms and a minimum of 250 ms, as the arms table under Consensus shows. This bundle is built from the records saved with the A3 blast, taken when the A4 blast began. In that copy arm A4 has a begin time (00:39:30 UTC) but no end, because it was still open; the run's final manifest closes A4 at 01:17:55 UTC. The A0 arm record was entered by hand because the apply tool records no arm when every node already holds the requested values, and A0 was the deployed configuration; its times are those of the deployment and the A1 apply.
  • Idle figures come from the window between the last node's confirmation of an arm's values and the start of that arm's funding step, about eight minutes. The funding step sends 6,000 transactions and is not idle. The per-arm idle and certification numbers are from the node logs; the Prometheus and machine metrics of this page are those saved with the A3 blast.
  • The comparison run, r20261006-0458, ran on five regions with two validators each, with a chaos kill of validator-05, 2,400 s of mixed load per generator and the generator build 3644adc; here the fleet is ten regions with one validator each, chaos is off, each blast is 480 s of mixed load and the generator build is 9f61303. The node build, image and the A0 timing are the same. Differences in throughput or latency between the two pages cannot be assigned to the timing arms.
  • Each arm was run once, in a fixed order on one deployment, so run-to-run variation and order effects are not measured.
  • The tag adiri:mmap-subsecond-certfix-10-6-2026 is mutable. The harness resolved it to digest sha256:caeb6be2… before any node was created, and every node ran that digest.
  • The nodes ran more than one max_header_delay_ms (250, 500, 800, 1000) by design: the fleet value is published here, and the per-node values are listed under consensus.param_arms.
  • The nodes ran more than one min_header_delay_ms (50, 250, 400, 500) by design: the fleet value is published here, and the per-node values are listed under consensus.param_arms.

Full report

Full report HTML · 1.3 MB

Changes from the original report:

  • The site adds a banner at the top of the report that links back to this page.

SHA-256 of the committed file, before the banner is added: 1811bc45d78be61e7b7fb599377556040a9d24b4c3f0d493605f468cdf96659a