What is the bandwidth-delay product?

The bandwidth-delay product (BDP) is the amount of data that can be in transit on a network path at any given moment. It equals the link bandwidth multiplied by the round-trip time. A TCP window smaller than the BDP will stall the sender while waiting for ACKs, reducing throughput.

Does the TCP window size need to be a power of two?

No. Modern TCP window scaling (RFC 7323) uses a bit-shift scale factor, allowing the advertised receive window to be any multiple of 2^S. The true optimal window size is the unrounded BDP. Sizing memory buffers to powers of two is an optional memory allocator optimization, not an OS or TCP protocol requirement.

How do I apply this on Linux?

Set sysctl net.core.rmem_max and net.ipv4.tcp_rmem max to at least 2 × BDP. By default (tcp_adv_win_scale = 1), the Linux kernel reserves about 50% of the socket buffer for sk_buff data structure overhead. Setting tcp_rmem to 2 × BDP ensures the advertised receive window reaches the full line-rate BDP.

What RTT should I use?

Use the RTT to your most distant or most performance-critical peer. Run "ping " and use the average value. For cross-continental paths expect 80–180 ms; for intercontinental 150–300 ms.

Does TCP window scaling change this?

TCP window scaling (RFC 7323, obsoleting RFC 1323) allows windows larger than 65,535 bytes by using a shift factor. Modern OSes negotiate it automatically. This calculator gives you the raw byte target; your OS handles the scaling negotiation.

TCP Window Size Calculator

Enter your link speed and round-trip time to calculate the optimal TCP window size that maximizes network throughput.

Link Parameters

Link Speed (Bandwidth)

Nominal line rate of your link (e.g. 1000 Mbps = 1 Gbps)

Round-Trip Time (RTT)

Measured ping RTT to the remote host in milliseconds

Current Window Size (optional)

Your current window to compare efficiency against optimal

Quick Presets

Enter link speed and RTT, then click Calculate.

Bandwidth-Delay Product (BDP)

—

Bytes

—

Recommended Window Size

—

Bytes

—

Theoretical Max Throughput

—

Mbps

—

Gbps

Linux Sysctl Reference (2× BDP for Kernel Overhead)

# Sized to 2× BDP to account for default Linux sk_buff overhead (tcp_adv_win_scale = 1)

net.core.rmem_max = —

net.core.wmem_max = —

# TCP socket buffers: min default max

net.ipv4.tcp_rmem = 4096 87380 —

net.ipv4.tcp_wmem = 4096 87380 —

Copied!

Summary

The TCP window size determines how much data can be in-flight on the network before an acknowledgment is required. Too small a window throttles throughput; the ideal size equals the bandwidth-delay product (BDP) — the amount of data that fills the pipe end-to-end. This calculator takes your link speed and round-trip time, computes the optimal TCP window size, and provides tuned Linux socket buffer (sysctl) settings accounting for kernel overhead.

How it works

Enter your link speed in Mbps or Gbps.
Enter the round-trip time (RTT) in milliseconds.
The calculator computes the bandwidth-delay product: BDP = Bandwidth x RTT.
The optimal receive window equals the BDP, while Linux sysctl buffers are scaled to 2× BDP for kernel sk_buff overhead.
Results are shown in bytes, kilobytes, and megabytes for easy reference.
A throughput efficiency estimate shows the impact of a smaller window.

Use cases

Tune Linux tcp_rmem / tcp_wmem sysctl parameters for high-speed WAN links.
Size the TCP receive window for satellite or long-haul fiber connections.
Diagnose slow file transfers where the window is the bottleneck.
Plan buffer sizes for network appliances and load balancers.
Validate iperf3 --window flag values before benchmarking.
Educate network engineers on the BDP concept with real numbers.

Frequently Asked Questions

Last updated: 2026-06-11 · Reviewed by Nham Vu