Concept 1 of 71 min
Latency budgets, colocation, fibre and microwave
Software against an FPGA
| Edge lasts | 40 µs |
|---|---|
| Network | 15 µs |
| Software | 25 µs |
| FPGA | 4 µs |
- Software = 25 + 15 = 40 µs, no margin
- FPGA = 4 + 15 = 19 µs
Only the FPGA (field-programmable gate array) acts while the edge is still there.
Fibre against radio
| Distance | 340 km |
|---|---|
| Fibre | 200,000 km/s |
| Radio | 300,000 km/s |
- Fibre = 340 ÷ 200,000 = 0.0017 s
- Radio = 340 ÷ 300,000 ≈ 0.00113 s
Radio arrives in about 1.13 ms against 1.7 ms, a lead worth paying for.
They must fit it with a margin. So delay = distance ÷ 200,000.
Colocation puts a firm's servers metres from the exchange's matching engine. Microwave links are faster because signals travel about a third faster through air on straighter paths.
A path that uses the whole window has no margin, and in a latency race the margin is the whole game. Fibre routes are also longer than the straight line, which is why firms built microwave links between centres such as Chicago and New Jersey, or London and Frankfurt.
Microwave carries less data and fades in bad weather, so it is used for the most time-sensitive messages only.
- Calling a path that uses the whole window fast enough.
- Using the speed of light in air for fibre.
- Thinking microwave carries more data than fibre.
Why would a trading firm pay a lot for a link that is only a millisecond faster than its competitor's?
A complete answer covers
- In arbitrage, the first order to arrive gets the trade.
- A millisecond is many times the time it takes to process an order.
- Being second most of the time earns nothing or loses money.
- So speed is bought where it decides who wins: links, colocation, hardware.
Also in our technical interviews, on camera, with a score.
Concept 2 of 71 min
Inside a low-latency engine
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 3 of 71 min
Orders, safety nets and the road to production
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 4 of 71 min
Uptime, change control and migrations
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 5 of 71 min
Core systems: mainframes, batches, messages and grids
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 6 of 71 min
EU rules bank engineers work under
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inConcept 7 of 71 min
AdvancedWhy latency explodes near full load
The why, the charts, the worked example and the interview questions are in Essentials, with every quiz.
See the plansSign inRecap
- A latency edge is a budget: processing + network must fit inside it with a margin; light in fibre covers about 200 km per millisecond.
- Trading engines preallocate memory, bypass the kernel and are judged on tail latency, not the average.
- Every order passes pre-trade checks; kill switches and limits cap the damage of any bug; algorithms go from backtest to simulation to small size.
- Uptime is protected by reviewed changes with a rollback plan, freezes around critical runs and gradual migrations.
- EU rules: clocks within 100 µs of UTC (Coordinated Universal Time) for high-frequency trading, PSD2 account access with consent, DORA for digital resilience, BCBS 239 (Basel Committee on Banking Supervision) for risk data.
25 questions on Technology in banks and trading
Timed, like the real one, drawn from a bank of 33. Every answer explained.