Throughput

Throughput is the actual amount of data successfully transferred per second under real network conditions. It is what a speed test measures. Throughput is always lower than bandwidth (the theoretical maximum) because of protocol overhead, congestion, and hardware limitations.

Throughput vs bandwidth vs speed

Bandwidth is the theoretical maximum capacity of a link — the size of the pipe. Your ISP sells you a bandwidth tier: "up to 500 Mbps." Throughput is the actual data rate you achieve at a given moment under real conditions — always equal to or less than bandwidth. Speed is the colloquial term people use interchangeably with throughput; in a speed test context they mean the same thing. The distinction matters when troubleshooting: if your 500 Mbps plan delivers 120 Mbps on a wired test, the gap between bandwidth (500) and throughput (120) tells you something is wrong in the path between you and the test server.

Why throughput is always less than bandwidth

• Protocol overhead: Every TCP/IP packet carries headers — IP header (20 bytes), TCP header (20 bytes minimum), Ethernet frame overhead — that consume capacity without delivering user data. At typical packet sizes, this overhead is 2–5% of total capacity.
• TCP flow control and slow start: TCP begins each new connection in a slow-start phase, gradually increasing the sending rate until it reaches the available capacity. Short transfers (small files, web page resources) may never exit slow start, delivering far below theoretical bandwidth.
• Retransmissions: Any lost packet must be retransmitted, consuming bandwidth a second time while adding delay. On links with 1% packet loss, retransmissions plus the congestion-window reduction TCP applies can reduce effective throughput by 20–30%.
• Congestion: Shared cable HFC nodes, congested ISP backbone links, or overloaded peering points all reduce available capacity below the provisioned plan speed.
• Wi-Fi overhead: Wi-Fi is half-duplex at the radio layer — a device cannot transmit and receive simultaneously on the same channel. Contention, retransmissions, and management frames reduce Wi-Fi throughput to 50–70% of the link's raw data rate.
• Hardware bottlenecks: An underpowered router CPU processing NAT and firewall rules, an old modem, or a NIC driver running in software mode can all cap throughput below the WAN connection speed.
• Server-side limits: The remote server may be bandwidth-capped, geographically distant, or under heavy load — none of which are related to your local connection.

TCP throughput limits at high latency (BDP)

TCP's congestion window limits the amount of unacknowledged data in flight at any moment. The Bandwidth-Delay Product (BDP) defines the maximum throughput achievable: BDP = bandwidth × round-trip time. On a 1 Gbps link with 100 ms RTT, the BDP is 12.5 MB — meaning TCP must keep 12.5 MB of data in flight to fully utilise the link. If the TCP receive window is smaller than the BDP (common on older operating systems with default socket buffer sizes), throughput will be throttled well below the available bandwidth. This is why high-bandwidth, high-latency links — such as satellite internet — are particularly difficult to fully saturate with a single TCP stream, even when the raw bandwidth is ample.

Throughput vs goodput

Goodput is a more precise metric than throughput: it counts only the payload bytes that are successfully delivered to the application, excluding retransmitted packets, duplicate ACKs, and protocol headers. Throughput includes all bits on the wire; goodput excludes the overhead. A speed test measures something close to goodput — the test server measures how many bytes the client successfully received, not the total bits transmitted including retransmissions. In practice, on a healthy connection with low packet loss, the difference between throughput and goodput is small (under 5%). On a lossy connection, goodput can be significantly lower than raw throughput.

Factors that reduce throughput

• Packet loss: Even 0.1% sustained packet loss noticeably degrades TCP throughput due to retransmissions and window reduction
• Jitter: Variable delay causes TCP's receive buffer to stall waiting for out-of-order packets, reducing effective throughput
• Distance / high RTT: High round-trip time reduces the rate at which TCP can grow its congestion window, limiting throughput on a single stream
• Bufferbloat: Oversized router buffers introduce queuing delay that mimics high latency, degrading interactive application performance even when raw throughput appears fine

How throughput is measured

Speed tests (SpeedTestHQ, Ookla, fast.com) measure throughput by downloading and uploading test files between your device and a nearby server, reporting the average data rate in Mbps. They use multiple parallel TCP streams to overcome single-stream TCP window limits and give a result close to real-world sustained download performance. iPerf3 is the standard tool for measuring throughput between two specific endpoints — useful for diagnosing LAN performance, VPN overhead, or ISP path characteristics. Run iperf3 -s on the server and iperf3 -c <server-ip> on the client. iPerf3 can also test UDP throughput, which is useful for measuring capacity without TCP's congestion control constraints.

Measuring throughput on a home network

For an accurate WAN throughput measurement: connect a device directly to your router via Ethernet, close background applications, and run a speed test to a server geographically close to you. Test at multiple times of day — morning and late evening — to separate peak-hour congestion from hardware limits. If wired throughput is significantly below your plan tier at off-peak hours, the issue is likely in the modem, router, or the ISP's provisioning. If throughput is fine off-peak but degrades in the evening, the shared cable node is congested.

What is a good throughput-to-bandwidth ratio?

On wired fiber, expect 90–100% of plan speed. On wired cable, 75–95%. On Wi-Fi (5 GHz, close to router), 60–85%. If your wired Ethernet throughput is consistently below 80% of your plan tier, there is a problem worth diagnosing.

Frequently Asked Questions

What is the difference between throughput and speed?

In networking, throughput is the precise technical term for what a speed test measures. 'Speed' is the colloquial version. They refer to the same thing: actual data transfer rate in Mbps, measured under real conditions.

Why is my throughput lower than my plan?

The most common causes are Wi-Fi (switch to Ethernet to eliminate this), an old router or modem that cannot process traffic fast enough, or peak-hour congestion on a shared cable node. Run a wired test at different times of day to isolate the cause.