Why 1000 Mbps Isn’t Just 125 MB/s: Understanding Network Overhead and MTU

Decimal and binary prefixes

Network speeds are expressed in decimal megabits per second (Mbps). Gigabit Ethernet (GbE) uses the decimal prefix “Giga”, so 1 GbE equals 1 000 000 000 bits / s. Disk capacities are usually given in decimal GB/MB, while memory sizes use binary GiB/MiB.

Conversion of 1000 Mbps to bytes

Because 1 Byte = 8 bits, a 1000 Mbps link can theoretically transfer 125 MegaByte / s (1000 000 000 / 8).

Ethernet frame overhead

An Ethernet frame consists of:

7‑byte preamble

Payload up to 1500 bytes

4‑byte Frame Check Sequence

The total size is 1500 + 38 = 1538 bytes, giving a raw efficiency of 1500 / 1538 ≈ 97.53 %.

If a 4‑byte VLAN tag is present, the frame size becomes 1500 + 38 + 4 = 1542 bytes, and efficiency drops to 1500 / 1542 ≈ 97.28 %.

Resulting payload throughput

Applying the efficiencies to the 1000 Mbps raw rate yields:

Without VLAN: 1000 Mbps × 0.9753 ≈ 975.3 Mbps (≈ 121.9 MB/s)

With VLAN: 1000 Mbps × 0.9728 ≈ 972.8 Mbps

File‑sharing protocols such as SMB add additional headers, so actual file transfer rates are lower than the payload throughput.

Jumbo frames

Increasing the MTU reduces per‑frame overhead. With a jumbo‑frame MTU of 9000 bytes, efficiency becomes 9000 / (9000 + 38) ≈ 99.58 %.

This raises the theoretical payload rate to 1000 Mbps × 0.9958 ≈ 995.8 Mbps, extracting more usable bandwidth for bulk transfers.

Trade‑offs of jumbo frames

Large frames are inefficient for small packets because a tiny payload still consumes an entire jumbo frame, wasting bandwidth. Consequently, networks often separate traffic:

Bulk data (e.g., storage, large file transfers) uses switches configured for jumbo frames.

Latency‑sensitive or small‑packet traffic uses standard‑MTU (1500 bytes) switches.

This separation aligns with storage‑oriented protocols such as iSCSI or Fibre Channel that run over Ethernet but have different performance requirements.

Code example

作者：
i
N在