Matched leader-paced A/B benchmark

Frankfurt DoubleZero TPU vs Public Internet

For each eligible leader run, the benchmark submitted one transaction through DoubleZero and one through Public Internet. Every performance delta below requires both transactions to be observed in the same Solana slot.

PoH placement: across 201 complete same-slot pairs, DoubleZero was earlier 91, both were in the same interval 106, and Public Internet was earlier 4. PoH placement describes produced-block order, not packet arrival at the leader TPU.
246paired leader runs
201same-slot pairs
29eligible leader identities
11.28%eligible scheduled slots

Overall same-slot comparison

Each delta is candidate minus baseline. Negative values favor the candidate. Percentiles select observed values without interpolation.

MeasurePairsDoubleZero earlierSamePublic Internet earlierDelta p50Delta p95Delta p99
First observation time201115086-0.006 ms0.018 ms4.408 ms
Processed transaction index201115086-2420
Entry index201113484-236
Preceding PoH ticks201911064001

Coverage outside the comparison population: 3 different-slot pairs, 2 baseline-only, 34 candidate-only, and 6 observed through neither route. These outcomes are reported as coverage and are never converted into timing or placement deltas.

By leader city

Cities and countries come from the getLeaderSlots snapshots saved during the run.

CohortValidatorsSame-slot nObservation DoubleZero/same/Public InternetObservation p50 / p95 / p99Tx index DoubleZero/same/Public InternetTx index p50 / p95 / p99Entry DoubleZero/same/Public InternetEntry p50 / p95 / p99PoH DoubleZero/same/Public InternetPoH p50 / p95 / p99
Beauharnois, CA17371 / 0 / 2-5.395 ms / -0.004 ms / 31.417 ms71 / 0 / 2-23 / -5 / 3671 / 1 / 1-22 / -6 / 3571 / 1 / 1-3 / -1 / 4
London, GB2560 / 0 / 560.009 ms / 0.112 ms / 4.408 ms0 / 0 / 561 / 4 / 70 / 0 / 561 / 4 / 70 / 56 / 00 / 0 / 0
Luxembourg, LU22817 / 0 / 11-0.005 ms / 0.023 ms / 7.263 ms17 / 0 / 11-1 / 5 / 517 / 0 / 11-1 / 4 / 54 / 24 / 00 / 0 / 0
Amsterdam, NL5126 / 0 / 60.005 ms / 0.013 ms / 0.013 ms6 / 0 / 61 / 7 / 74 / 3 / 50 / 5 / 50 / 11 / 10 / 1 / 1
Aubervilliers, FR1113 / 0 / 80.011 ms / 0.574 ms / 0.574 ms3 / 0 / 82 / 35 / 353 / 0 / 82 / 6 / 60 / 9 / 20 / 1 / 1
Singapore, SG11111 / 0 / 0-0.009 ms / -0.005 ms / -0.005 ms11 / 0 / 0-7 / -2 / -211 / 0 / 0-8 / -3 / -310 / 1 / 0-1 / 0 / 0
São Paulo, BR222 / 0 / 0-17.509 ms / -17.509 ms / -17.509 ms2 / 0 / 0-40 / -40 / -402 / 0 / 0-45 / -45 / -452 / 0 / 0-5 / -5 / -5
Tokyo, JP122 / 0 / 0-20.209 ms / -20.209 ms / -20.209 ms2 / 0 / 0-54 / -54 / -542 / 0 / 0-48 / -48 / -482 / 0 / 0-5 / -5 / -5
Bloomfield, US110 / 0 / 10.011 ms / 0.011 ms / 0.011 ms0 / 0 / 12 / 2 / 20 / 0 / 12 / 2 / 20 / 1 / 00 / 0 / 0
Hong Kong, HK110 / 0 / 10.005 ms / 0.005 ms / 0.005 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
London, EN111 / 0 / 0-0.011 ms / -0.011 ms / -0.011 ms1 / 0 / 0-17 / -17 / -171 / 0 / 0-10 / -10 / -101 / 0 / 0-2 / -2 / -2
Los Angeles, US111 / 0 / 0-0.009 ms / -0.009 ms / -0.009 ms1 / 0 / 0-3 / -3 / -31 / 0 / 0-4 / -4 / -41 / 0 / 0-1 / -1 / -1
Prague, CZ110 / 0 / 10.011 ms / 0.011 ms / 0.011 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
Stockholm, SE111 / 0 / 0-0.013 ms / -0.013 ms / -0.013 ms1 / 0 / 0-1 / -1 / -11 / 0 / 0-1 / -1 / -10 / 1 / 00 / 0 / 0

Outcome triplets are candidate earlier / same / baseline earlier. Every percentile uses only complete same-slot values for that measure.

By distance from sender

Distance is calculated from the saved validator coordinates to the sender coordinates supplied to the report command. Missing coordinates remain an explicit unavailable cohort.

CohortValidatorsSame-slot nObservation DoubleZero/same/Public InternetObservation p50 / p95 / p99Tx index DoubleZero/same/Public InternetTx index p50 / p95 / p99Entry DoubleZero/same/Public InternetEntry p50 / p95 / p99PoH DoubleZero/same/Public InternetPoH p50 / p95 / p99
Under 1,000 km1210927 / 0 / 820.008 ms / 0.052 ms / 4.408 ms27 / 0 / 821 / 5 / 2025 / 3 / 811 / 5 / 65 / 101 / 30 / 0 / 1
At least 5,000 km79086 / 0 / 4-0.312 ms / -0.004 ms / 31.417 ms86 / 0 / 4-22 / -2 / 3686 / 1 / 3-19 / -3 / 3585 / 4 / 1-2 / 0 / 4
1,000-5,000 km111 / 0 / 0-0.013 ms / -0.013 ms / -0.013 ms1 / 0 / 0-1 / -1 / -11 / 0 / 0-1 / -1 / -10 / 1 / 00 / 0 / 0
Coordinate unavailable111 / 0 / 0-0.009 ms / -0.009 ms / -0.009 ms1 / 0 / 0-3 / -3 / -31 / 0 / 0-4 / -4 / -41 / 0 / 0-1 / -1 / -1

Outcome triplets are candidate earlier / same / baseline earlier. Every percentile uses only complete same-slot values for that measure.

By validator identity

Each validator remains separate. Repeated leader runs from one validator are not independent validator samples; small cohorts are descriptive only.

CohortValidatorsSame-slot nObservation DoubleZero/same/Public InternetObservation p50 / p95 / p99Tx index DoubleZero/same/Public InternetTx index p50 / p95 / p99Entry DoubleZero/same/Public InternetEntry p50 / p95 / p99PoH DoubleZero/same/Public InternetPoH p50 / p95 / p99
Fd7btgySsrju...
Beauharnois, CA
17371 / 0 / 2-5.395 ms / -0.004 ms / 31.417 ms71 / 0 / 2-23 / -5 / 3671 / 1 / 1-22 / -6 / 3571 / 1 / 1-3 / -1 / 4
9eGrDohdNTAo...
London, GB
1310 / 0 / 310.009 ms / 0.260 ms / 4.408 ms0 / 0 / 311 / 4 / 50 / 0 / 311 / 4 / 50 / 31 / 00 / 0 / 0
5pPRHniefFjk...
Luxembourg, LU
12716 / 0 / 11-0.005 ms / 0.023 ms / 7.263 ms16 / 0 / 11-1 / 5 / 516 / 0 / 11-1 / 4 / 54 / 23 / 00 / 0 / 0
JD549HsbJHeE...
London, GB
1250 / 0 / 250.009 ms / 0.052 ms / 0.112 ms0 / 0 / 251 / 3 / 70 / 0 / 251 / 3 / 70 / 25 / 00 / 0 / 0
CoG8d9Fp2TFJ...
Singapore, SG
11111 / 0 / 0-0.009 ms / -0.005 ms / -0.005 ms11 / 0 / 0-7 / -2 / -211 / 0 / 0-8 / -3 / -310 / 1 / 0-1 / 0 / 0
ana2y2YvQ3ZP...
Aubervilliers, FR
1113 / 0 / 80.011 ms / 0.574 ms / 0.574 ms3 / 0 / 82 / 35 / 353 / 0 / 82 / 6 / 60 / 9 / 20 / 1 / 1
7cVfgArCheMR...
Amsterdam, NL
174 / 0 / 3-0.009 ms / 0.012 ms / 0.012 ms4 / 0 / 3-1 / 7 / 72 / 3 / 20 / 5 / 50 / 6 / 10 / 1 / 1
HH5dA42XF1Hx...
Amsterdam, NL
122 / 0 / 0-0.005 ms / -0.005 ms / -0.005 ms2 / 0 / 0-1 / -1 / -12 / 0 / 0-1 / -1 / -10 / 2 / 00 / 0 / 0
mds4GEuiSgQR...
Tokyo, JP
122 / 0 / 0-20.209 ms / -20.209 ms / -20.209 ms2 / 0 / 0-54 / -54 / -542 / 0 / 0-48 / -48 / -482 / 0 / 0-5 / -5 / -5
2dfgsiSaZ51Q...
Amsterdam, NL
110 / 0 / 10.005 ms / 0.005 ms / 0.005 ms0 / 0 / 13 / 3 / 30 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
2mMGsb5uy1Q4...
Stockholm, SE
111 / 0 / 0-0.013 ms / -0.013 ms / -0.013 ms1 / 0 / 0-1 / -1 / -11 / 0 / 0-1 / -1 / -10 / 1 / 00 / 0 / 0
ACvL73V4GNnx...
Luxembourg, LU
111 / 0 / 0-0.004 ms / -0.004 ms / -0.004 ms1 / 0 / 0-1 / -1 / -11 / 0 / 0-1 / -1 / -10 / 1 / 00 / 0 / 0
EBk678aQvc3c...
Los Angeles, US
111 / 0 / 0-0.009 ms / -0.009 ms / -0.009 ms1 / 0 / 0-3 / -3 / -31 / 0 / 0-4 / -4 / -41 / 0 / 0-1 / -1 / -1
ECZx4Dfyn2o5...
Bloomfield, US
110 / 0 / 10.011 ms / 0.011 ms / 0.011 ms0 / 0 / 12 / 2 / 20 / 0 / 12 / 2 / 20 / 1 / 00 / 0 / 0
H8fHToVcZPi5...
London, EN
111 / 0 / 0-0.011 ms / -0.011 ms / -0.011 ms1 / 0 / 0-17 / -17 / -171 / 0 / 0-10 / -10 / -101 / 0 / 0-2 / -2 / -2
HW4zorvt6xDw...
Hong Kong, HK
110 / 0 / 10.005 ms / 0.005 ms / 0.005 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
RNXnAJV1DeBt...
Prague, CZ
110 / 0 / 10.011 ms / 0.011 ms / 0.011 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
SA1LFXr4os2P...
São Paulo, BR
111 / 0 / 0-26.560 ms / -26.560 ms / -26.560 ms1 / 0 / 0-40 / -40 / -401 / 0 / 0-45 / -45 / -451 / 0 / 0-5 / -5 / -5
SSmBEooM7Rkm...
Amsterdam, NL
110 / 0 / 10.013 ms / 0.013 ms / 0.013 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
kom1oNHyyt84...
Amsterdam, NL
110 / 0 / 10.011 ms / 0.011 ms / 0.011 ms0 / 0 / 11 / 1 / 10 / 0 / 11 / 1 / 10 / 1 / 00 / 0 / 0
mALL2W6DUgDD...
São Paulo, BR
111 / 0 / 0-17.509 ms / -17.509 ms / -17.509 ms1 / 0 / 0-111 / -111 / -1111 / 0 / 0-77 / -77 / -771 / 0 / 0-5 / -5 / -5

Outcome triplets are candidate earlier / same / baseline earlier. Every percentile uses only complete same-slot values for that measure.

Methodology

Paired design

Both transactions share the leader run, slot trigger, sender host, transaction shape, and fee configuration. Only the network route differs.

Same-slot rule

First-observation time, processed transaction index, entry index, and PoH interval are compared only when both signatures land in the same slot.

Observation clock

Each signature uses its earliest event across the configured RPCEdge receivers. Deshred and processed receivers are not compared with each other.

Block placement

Processed transaction index is canonical transaction order. Entry index and preceding PoH ticks are derived after collection from the saved contiguous entry history.

Interpretation

Observation time includes sender scheduling, network transit, leader admission, block production, and propagation to the Frankfurt observer. It is not network RTT or TPU packet-arrival latency.

Geography

City, country, latitude, and longitude are historical values saved in getLeaderSlots snapshots. The coarse region classifier is not used.

The report does not claim that repeated transactions from one validator are independent samples or that geographic distance alone caused an observed difference.