Concepts and theorems · Intermediate

Bandwidth-delay product: how much data must be in flight

Compute an in-flight data budget from bandwidth and RTT, then distinguish receive-window limits from congestion limits.

Ask Logeia about this articleTry the three questionsReview your answers

Updated

On this page
  1. Multiply a rate by a round-trip duration
  2. Calculate a concrete window budget
  3. Identify all the limits on outstanding data
  4. Keep the model within its scope
  5. Ask Logeia about this articleCheck your understanding

Multiply a rate by a round-trip duration

Bandwidth-delay product, or BDP, is a quantity of data obtained by multiplying a desired transmission rate by a round-trip duration. With bits per second and seconds, the result is bits. In TCP discussions, it helps estimate how much unacknowledged data must be allowed in flight to keep sending while feedback travels back. Be explicit about whether the rate is nominal link capacity or an achievable share of a path, and whether the delay is a round trip. Mixing one-way delay, RTT, bits, and bytes can change both the result and its meaning.

Calculate a concrete window budget

Use 100 Mbps as 100,000,000 bits per second and forty milliseconds as 0.04 seconds. Their product is 4,000,000 bits, or 500,000 bytes: about 500 kB in decimal units. If a simplified model permits only 64,000 bytes outstanding, its window-derived rate ceiling is 64,000 divided by 0.04, or 1,600,000 bytes per second, approximately 12.8 Mbps. A nominal 100 Mbps path therefore does not guarantee that one connection with that outstanding-data limit can fill it. The calculation isolates a possible constraint rather than measuring actual performance.

QuantityValue
Rate100,000,000 bit/s
RTT0.04 s
Product4,000,000 bit
Convert to bytes500,000 byte

Identify all the limits on outstanding data

TCP’s receiver-advertised window and the sender’s congestion-control window both affect sending, along with data already in flight and application behavior. Enlarging a receive buffer alone cannot overcome a small congestion window, an application that supplies too little data, or a receiver that is not reading. Window scaling allows larger receive windows to be represented; it does not reserve bandwidth or eliminate packet loss. Use BDP to formulate a testable explanation, then inspect the relevant windows and transfer behavior instead of assuming that one buffer adjustment controls the entire connection.

Keep the model within its scope

BDP is useful for reasoning about sustained transfer, not as a complete prediction for the completion time of one small request. Startup behavior, retransmission, competing traffic, CPU work, and storage can all affect measured throughput. An RTT inflated by queueing can also make naïvely increasing buffers counterproductive by increasing waiting further. Record the units and assumptions of the calculation, then compare transfer volume, timing distribution, and window observations. A sensible diagnostic question is whether the amount permitted in flight is limiting the workload; it is not whether every buffer should always be set to one universal BDP number.

Reading and practice are recorded separately in this browser, for each language.

Learn this in order

Understand the systems behind the web · STEP 10

Choose a learning path