Bandwidth-Delay Product (BDP) & TCP Window Scaling in High-Speed Networks

1. The Fundamental Problem: Why High-Bandwidth Links Feel Slow

A common operational grievance reported by network engineers is purchasing a high-speed 10 Gbps or 100 Gbps dedicated IP transit circuit, only to find that a single TCP file transfer (e.g., SCP, FTP, or HTTPS) between data centers in New York and London crawls along at less than 50 Mbps. Running an iPerf test with multiple parallel streams achieves full line rate, yet a single thread stalls.

This is not a physical hardware failure. It is the direct consequence of the Bandwidth-Delay Product (BDP) and the sliding window flow control mechanics governing the Transmission Control Protocol (TCP).

2. Defining the Bandwidth-Delay Product

The Bandwidth-Delay Product (BDP) represents the total volume of data that can be "in flight" across a network path at any single instant in time. It is analogous to the total volume of water contained within a pipeline of a given cross-sectional diameter (bandwidth) and length (round-trip time):

To keep the 10 Gbps pipeline 100% full without stalling, the sender must transmit 95.37 MB of unacknowledged data before pausing to wait for the first TCP ACK packet to return across the Atlantic.

3. The 16-Bit Window Limit and RFC 7323 Window Scaling

In the original TCP specification (RFC 793, written in 1981), the Receive Window (RWIN) field in the TCP header was allocated exactly 16 bits. The maximum buffer size that a receiver could advertise to a sender was 2¹⁶ − 1 = 65,535 bytes (64 KB).

If a sender is constrained by a 65,535-byte window on our 80 ms transatlantic link, the maximum throughput it can mathematically achieve—regardless of whether the underlying circuit is 1 Gbps, 10 Gbps, or 100 Gbps—is governed by the window flow ceiling:

To overcome this crippling limitation on modern high-speed broadband, the IETF created RFC 7323 (TCP Extensions for High Performance), defining the Window Scale Option. During the initial TCP three-way handshake (SYN, SYN-ACK), endpoints negotiate a scaling multiplier between 0 and 14, shifting the 16-bit window up to 30 bits, enabling maximum window sizes of up to 1 Gigabyte (1,073,725,440 bytes) and allowing single-threaded TCP flows to easily saturate 100 Gbps links.