Rate & Flow Estimation

Medium2–3 hrsRateFlowLittle's LawThroughput

Overview

Rate and flow problems require you to reason about steady-state quantities using Little's Law (L=λW) and bottleneck analysis. In trading contexts, this appears when estimating market impact (how quickly can you trade X shares?), order flow (how many messages does an exchange handle per second?), or hardware throughput (can our co-location server handle the tick data volume?). The key tools are Little's Law, unit conversion chains, and identifying the tightest constraint.

How to Recognize

→'How fast does X happen?'
→'How many Y happen per second/day/year?'
→Queue length and wait time problems
→'How long does it take to...'
→Any question about throughput or capacity

Step-by-Step Approach

1.Use Little's Law: L = λW (queue length = arrival rate × wait time)
2.Identify the bottleneck — the rate-limiting step determines the whole system rate
3.Anchor on a known rate (heartbeats/min, typing speed, internet bandwidth)
4.Convert units carefully: sec → min → hour → day → year
5.For economies of scale: double the throughput, roughly double the cost

Key Formulas

L = \lambda W

— Little's Law (avg queue length = arrival rate × avg wait)

\text{Throughput} = \min(\text{capacity of each step in pipeline})

1\text{ year} \approx \pi \times 10^7\text{ seconds}

(useful mnemonic)

\text{Heartbeats/lifetime} \approx 70 \times 60 \times 24 \times 365 \times 80 \approx 3 \times 10^9

Worked Examples

0Quick Example

How fast can you transfer 1TB over the internet? Broadband ≈ 100 Mbps = 12.5 MB/s. 1TB = 10^6 MB. Time = 10^6/12.5 ≈ 80,000 sec ≈ 22 hours. With fiber (1 Gbps): ~2.2 hours.

Problem

Estimate the peak message rate (orders + cancels + trades) at the NYSE or NASDAQ on a typical high-volume day.

Solution

US equity market trades roughly 10B shares/day total across all exchanges. NYSE + NASDAQ handle roughly half: ~5B shares.

Average trade size: ~200 shares. Trades executed: 5B/200 = 25M trades/day on the two main exchanges.

For each executed trade, there are roughly 20–50 order/cancel messages (most orders are cancelled). Messages ≈ 25M × 30 = 750M messages/day.

Trading hours: 6.5 hours = 23,400 seconds. Peak hours (first/last 30 min) see 3× average. Avg rate: 750M/23400 ≈ 32k msg/sec. Peak: ~100k msg/sec.

Actual: NASDAQ processes ~500k–1M messages/second at peak. We're off by ~5–10×, but the right order of magnitude.

Answer

Peak rate: ~100k–1M messages/second. Our estimate (100k) is a lower bound; actual systems handle ~10× more due to cancellations and quote updates.

Common Mistakes

!
Forgetting that 1 year ≈ 3×10^7 seconds (not 10^8 or 10^6). Quick check: 365 × 24 × 3600 ≈ 31.5M ≈ π×10^7.
!
Not identifying the bottleneck. The rate is always limited by the slowest step — find it first.
!
Confusing 'orders' and 'trades'. In equity markets, ~95% of orders are cancelled; executed trade rate is much lower than total message rate.

Practice Problems

Click "Show Answer" to reveal

Estimate the number of heartbeats in a human lifetime.

Hint: Average heart rate: ~70 bpm. Average lifespan: ~80 years. Convert years to minutes.

Estimate the bandwidth required to stream all trades and quotes from all US equity exchanges in real time.

Hint: ~10B shares/day traded, but quotes are 100–1000× more data than trades. Estimate number of quote updates per day, then data per message.

Back to Study Plan

The Fermi Method Sanity Checking Calculations