How to Benchmark Jito Latency Correctly

To benchmark Jito latency correctly, measure round-trip time from your candidate server to the specific Jito transaction sender you will use, continuously over time rather than once, and report the minimum, average, and maximum — not just the average. A single ping tells you almost nothing; the spread is what predicts real trading behaviour.

Measure to the right target

Latency is only meaningful against the endpoint you actually submit to. Benchmark to the Jito block engine / transaction sender in the region you target, from the exact datacenter you are evaluating — not from your laptop, and not to a generic host.

Measure continuously, not once

Network paths vary. A one-off ping can be misleadingly fast or slow. Run a sustained sample so you can see how the connection behaves over time:

ping -c 1000 <your-target-jito-endpoint>

Better still, use mtr to see the per-hop path and where latency or loss is introduced.

Report min, average, and max — and why

The average hides the tail. For trading and MEV, the worst case matters most, because that is when you lose. Always report the full spread: the minimum (your best case), the average (typical), and the maximum (worst case in the window). A consistent 1ms often beats a 0.5ms that regularly spikes.

This is exactly what we publish on our Jito latency benchmarks page: measured min/avg/max per location, with a machine-readable feed at /data/jito-latency.json. We publish only what we measure — we do not estimate percentiles or numbers we cannot stand behind.

Common mistakes

• Benchmarking from the wrong place — your home connection is not your server.
• Trusting a single sample — one fast ping is not a benchmark.
• Reporting only the average — it hides the spikes that cost you.
• Comparing apples to oranges — same target, same method, same window, or the comparison is meaningless.

FAQ

What latency to Jito is good?

It depends on your region and strategy, but sub-millisecond is achievable from a datacenter in the same metro with direct peering. See the live measured numbers on our benchmarks page.

Why don't you publish p95 / p99?

We publish exactly what we measure — minimum, average, and maximum. We would rather show fewer honest numbers than impressive ones we cannot stand behind.

Benchmark from our network

Spin up a low-latency VPS or bare metal in a Jito-connected location and run your own measurements — start at orbitservers.io.