TCP Window Size Calculator

MTU (Maximum Transmission Unit) is the largest packet size a network can transmit, typically 1500 bytes for Ethernet. MSS (Maximum Segment Size) is the maximum TCP payload size, calculated as: MSS = MTU - IP Header (20 bytes for IPv4, 40 for IPv6) - TCP Header (20 bytes minimum + options). With standard Ethernet and IPv4, MSS = 1500 - 20 - 20 - 12 (timestamps) = 1448 bytes.

For WAN links, your TCP receive buffer should be at least equal to the Bandwidth-Delay Product (BDP). For high-latency links (100ms+ RTT), multiply BDP by 2-4x for optimal performance. For example, a 100 Mbps link with 200ms RTT needs at least 2.5 MB buffers. The calculator generates OS-specific commands for Linux (sysctl), Windows (netsh), and macOS.

TCP window scaling (RFC 7323) is needed when the Bandwidth-Delay Product exceeds 64 KB (65,535 bytes), which is the maximum value in the standard 16-bit TCP window field. Any modern high-speed or long-distance link requires window scaling. The scale factor is negotiated during the TCP handshake and allows windows up to 1 GB. Most operating systems enable this by default, but it's important to verify on older systems.

For high-latency WAN links and satellite connections, BBR (Bottleneck Bandwidth and RTT) generally outperforms loss-based algorithms. For datacenter and low-latency environments, CUBIC (Linux default) works well. For networks with high packet loss, BBR is more resilient. The calculator recommends algorithms based on your specific network characteristics.

The Mathis formula estimates maximum throughput under packet loss: Throughput ≈ (MSS / RTT) × (1.22 / √loss_rate). With 1% packet loss, a link with 1500-byte MSS and 100ms RTT is limited to about 14.6 Mbps regardless of available bandwidth. This makes reducing packet loss critical for WAN performance. BBR congestion control handles loss better than traditional algorithms like CUBIC or Reno.

Linux TCP buffers are controlled by three sysctl parameters: net.core.rmem_max/wmem_max (maximum buffer size), net.ipv4.tcp_rmem/tcp_wmem (minimum, default, maximum per-socket), and net.core.netdev_max_backlog (packet queue size). For high-performance WAN links, set rmem_max/wmem_max to at least 2× your BDP, and tcp_rmem/tcp_wmem's maximum to the same value.

TCP throughput is limited by the smallest of three factors: available bandwidth, TCP window size / RTT, and Mathis limit (packet loss impact). If your window size is too small for your link's BDP, throughput will be capped. Run this calculator with your bandwidth, RTT, and packet loss to identify the bottleneck and get tuning recommendations.

Default TCP buffer sizes vary by OS: Linux typically defaults to 128 KB - 6 MB (auto-tuning enabled), Windows 10/11 defaults to 64 KB with auto-tuning, and macOS defaults to 256 KB - 4 MB. These defaults work for most LAN scenarios but may limit performance on high-bandwidth WAN links. The calculator shows current defaults and recommends optimal values for your network.

VPNs add encapsulation overhead that reduces effective MTU: IPsec typically uses 1400 bytes, WireGuard uses 1420 bytes, and OpenVPN uses 1450 bytes. If MTU is not adjusted, packets may fragment or be dropped, severely impacting performance. The MTU/MSS Optimizer tab helps calculate correct values for various tunnel types and generates PMTUD recommendations.