← Benchmarks

BENCHMARK · TX SENDER · PAIRED A/B

DoubleZero IBRL versus public Internet

One-hour leader-paced A/B from Frankfurt: for every eligible Solana leader window (DZ TPU open), one transaction through DoubleZero IBRL and one through the public Internet. Same host, relay binary, transaction shape, blockhash, slot trigger, and leader TPU endpoint.

2026-07-13 · 02:54 UTC1 hourDoubleZero IBRL vs public Internet · Frankfurtleader-paced paired A/B · mainnet

◆ HEADLINE · POH PLACEMENT

PoH earlier or tied

98.0%

197 of 201 same-slot pairs

PoH · DoubleZero earlier

91

106 same · 4 public earlier

Same-slot pairs

201

246 paired leader runs · 29 identities

Eligible leader windows

11.3%

scheduled slots where leader had DZ TPU open

How to read this. PoH placement and processed transaction index describe produced-block order, not packet arrival at the leader TPU. Observation time is a read-path signal that can move for reasons beyond the send route. Deltas are candidate minus baseline; negative favors DoubleZero. Comparisons use same-slot pairs only. Across 201 same-slot pairs, DoubleZero was earlier on preceding PoH ticks 91 times, both arms shared the interval 106 times, and Public Internet was earlier 4 times.

SAME-SLOT COMPARISON

Placement first. Observation second.

MeasurePairsDoubleZero earlierSamePublic Internet earlierΔ p50Δ p95Δ p99
Preceding PoH ticksprimary20191106400+1
Processed transaction indexprimary201115086-2+4+20
Entry index201113484-2+3+6
First observation time201115086-0.006 ms+0.018 ms+4.408 ms

Deltas are DoubleZero minus Public Internet. Negative values favor DoubleZero. Observation time can agree with placement while still moving for receiver/read-path reasons - use PoH ticks and processed index as the decision metrics.

BY DISTANCE FROM SENDER

The edge scales with leader distance.

Within Europe the PoH improvement is negligible - the public path almost never lands in an earlier PoH interval. Past ~5,000 km, DoubleZero leads on block placement in the large majority of pairs.

Preceding PoH ticks by distance
Distance cohortnValidatorsDoubleZero / same / Public InternetΔ p50Δ p95Δ p99
Under 1,000 km109125 / 101 / 300+1
At least 5,000 km90785 / 4 / 1-20+4
1,000-5,000 km110 / 1 / 0000
Coordinate unavailable111 / 0 / 0-1-1-1

Outcome triplet is DoubleZero earlier / same / Public Internet earlier. Negative deltas favor DoubleZero.

Processed transaction index by distance
Distance cohortnValidatorsDoubleZero / same / Public InternetΔ p50Δ p95Δ p99
Under 1,000 km1091227 / 0 / 82+1+5+20
At least 5,000 km90786 / 0 / 4-22-2+36
1,000-5,000 km111 / 0 / 0-1-1-1
Coordinate unavailable111 / 0 / 0-3-3-3

Outcome triplet is DoubleZero earlier / same / Public Internet earlier. Negative deltas favor DoubleZero.

BY LEADER CITY

Where the result concentrates.

Preceding PoH ticks by leader city
CitynValidatorsDoubleZero / same / Public InternetΔ p50Δ p95Δ p99
Beauharnois, CA73171 / 1 / 1-3-1+4
London, GB5620 / 56 / 0000
Luxembourg, LU2824 / 24 / 0000
Amsterdam, NL1250 / 11 / 10+1+1
Aubervilliers, FR1110 / 9 / 20+1+1
Singapore, SG11110 / 1 / 0-100
São Paulo, BR222 / 0 / 0-5-5-5
Tokyo, JP212 / 0 / 0-5-5-5
Bloomfield, US110 / 1 / 0000
Hong Kong, HK110 / 1 / 0000
London, EN111 / 0 / 0-2-2-2
Los Angeles, US111 / 0 / 0-1-1-1
Prague, CZ110 / 1 / 0000
Stockholm, SE110 / 1 / 0000

Outcome triplet is DoubleZero earlier / same / Public Internet earlier. Negative deltas favor DoubleZero.

Cities come from getLeaderSlots snapshots saved during the run. Small cohorts are descriptive only.

NOTES

The honest footnotes.

  • Eligible windows only: leader has DoubleZero TPU open. Both arms shared the Frankfurt host, relay binary, transaction shape, recent blockhash, slot trigger, and leader TPU endpoint.
  • DZ IBRL improvement scales with leader distance. Beyond 5,000 km, DoubleZero led on preceding PoH ticks in 85 of 90 same-slot pairs.
  • Within Europe (under 1,000 km) the PoH edge is negligible: 101 of 109 pairs landed in the same PoH interval; the public Internet path was earlier on PoH in only 3 pairs.
  • Observation-time deltas roughly track PoH and slot ordering but include receiver/read-path effects - treat placement metrics as the primary comparison.
  • Coverage outside the comparison population: 3 different-slot pairs, 2 baseline-only, 34 candidate-only, 6 observed on neither route. Those are coverage, not timing deltas.

// raw report: https://rpc-edge.github.io/solana-tx-bench/reports/leader-paced-2026-07-13T025442.353Z/