4,000 bikes riding at once. For 4 hours straight.
We filled our live platform with 4,000 connected bikes and kept them in rides for 4 hours, with riders unlocking, riding, pausing, locking and paying the whole time. There were no server errors and no lost location reports.
4,000
bikes connected and ridden at the same time
4 hours
held without a break
72,357
rides completed and paid for
10.7M
location and status reports received from the locks
0
server errors
0
location reports lost
What riders felt at full load
Times for 95 out of every 100 actions, measured from the rider's side across the public internet, with 4,000 bikes connected.
99.92% of commands sent to the locks were answered in time.
A test harsher than any real day
We set every condition to its worst case, so a real fleet's busiest day sits well inside what we measured.
Every bike in a ride at once
On a real fleet's busiest hour, a minority of bikes are out on rides at the same time. The rest are parked and quiet.
A position every 5 seconds
From every bike in a ride, for the whole test.
Every rider's app refreshing every 5 seconds
Real riders put their phones away while they ride.
4 hours without a break
Real demand rises and falls through the day.
5 operator consoles watching live
Polling the live map and fleet the whole time.
Over the open internet
Traffic came from a separate site, with ordinary internet delays, not from inside a lab.
What it means for your fleet
Even with all 100 bikes out at once, your fleet would use 2.5% of the load we proved, which is 40 times the room you need. On an ordinary day most bikes are parked, which leaves even more.
How the test was run
Real lock protocol. Each simulated lock spoke the same protocol as the hardware locks on our fleets: it connected, reported its position and status, and answered unlock and lock commands.
Real rides end to end. Simulated riders opened the app, checked prices, unlocked, rode for 5 to 20 minutes, paused, locked and paid from their wallets.
Our live platform. The test ran on the production platform operators use, not on a separate test system.
Graded fairly. Results count from once the fleet was connected and riding. Minutes when the testing site's own internet connection dropped were left out, because they say nothing about the platform.