Network Effective Throughput & Goodput Calculator

1. Wire Rate vs. Throughput vs. Goodput: The Network Hierarchy

In modern computer systems and carrier telecommunications, performance metrics are frequently conflated across distinct layers of the Open Systems Interconnection (OSI) reference model. Network operators and cloud engineers must rigorously distinguish between four fundamental throughput tiers:

• Physical Layer (L1) Wire Rate: The raw frequency at which a physical network interface controller (NIC) and optical transceiver clock serialized binary bits across the transmission medium. For a standard 10GBASE-LR optical link, the physical signaling clock asserts exactly 10,000,000,000 bits per second (10.000 Gbps). Every serialized symbol—including inter-packet gaps, preambles, frame headers, and error check sequences—consumes this budget.
• Data Link Layer (L2) Throughput: The volume of Ethernet frames traversing the Media Access Control (MAC) interface per second, excluding Layer 1 inter-packet gaps, preambles, and Start Frame Delimiters. L2 throughput measures total Ethernet payload capacity (including Layer 3 headers).
• Network Layer (L3) Throughput: The volume of Internet Protocol (IPv4 or IPv6) datagrams routed across network nodes, excluding Ethernet MAC address headers (14 Bytes), 802.1Q tags (4 Bytes), and cyclic redundancy checks (FCS - 4 Bytes). This corresponds to standard IP-level transit billing metrics.
• Application Layer (L7) Goodput: The net volume of usable application payload delivered into user space memory buffers (e.g., HTTP response bodies, database query results, or storage block payloads). Goodput strictly excludes all transport (TCP/UDP), network (IP), data link (Ethernet), physical (IPG), and overlay tunneling headers. Goodput represents the only bandwidth metric directly perceived by end users and application software.

2. The Anatomy of Layer 1 & Layer 2 Ethernet Overhead

Standard Ethernet communication (IEEE 802.3) enforces a mandatory physical and data link envelope around every transmitted frame. Even when transferring empty packets, this fixed framing overhead cannot be circumvented:

• Preamble (7 Bytes): A 56-bit sequence of alternating binary values (10101010...) that allows the receiver's physical layer (PHY) phase-locked loop (PLL) circuitry to achieve bit synchronization with the incoming signal clock.
• Start Frame Delimiter (SFD - 1 Byte): The exact bit pattern 10101011, which signals the physical receiver that clock synchronization is complete and the subsequent octet begins the destination MAC address.
• Inter-Packet Gap (IPG - 12 Bytes / 96 Bit-Times): A mandatory period of silence asserted between consecutive frames. At 10 Gbps, 96 bit-times equal exactly 9.600 nanoseconds. The IPG provides receiver transceivers with sufficient recovery time to process the completed frame, reset internal buffer pointers, and prepare for the subsequent frame preamble.
• Ethernet MAC Header (14 Bytes): Composed of a 6-byte Destination MAC address, a 6-byte Source MAC address, and a 2-byte EtherType field (e.g., 0x0800 for IPv4, 0x86DD for IPv6).
• Frame Check Sequence (FCS - 4 Bytes): A 32-bit cyclic redundancy check (CRC-32) calculated over the MAC header and payload to detect corrupted frames on the wire. Corrupted frames are silently dropped by hardware MAC controllers.

Together, the Layer 1 overhead (7B Preamble + 1B SFD + 12B IPG = 20 Bytes) and Layer 2 overhead (14B MAC + 4B FCS = 18 Bytes) enforce a fixed baseline penalty of 38 Bytes on every standard Ethernet packet before considering VLAN tags, IP, or TCP headers.

3. The Mathematical Case for Jumbo Frames in the Data Center

In high-performance compute clusters, storage area networks (iSCSI, NVMe-oF), and big-data fabrics, standard 1500-byte MTUs impose significant computational and bandwidth penalties. Jumbo frames (standardized at 9,000 Bytes MTU) dramatically alter the mathematical efficiency ratio:

• Standard 1500-Byte Frame:
  Total Wire Size = 1500 MTU + 18 L2 + 20 L1 = 1538 Bytes (12,304 bits).
  Usable TCP Payload (MSS) = 1500 − 20 (IPv4) − 20 (TCP) = 1460 Bytes.
  Efficiency η = 1460 ÷ 1538 = 94.93%.
• Jumbo 9000-Byte Frame:
  Total Wire Size = 9000 MTU + 18 L2 + 20 L1 = 9038 Bytes (72,304 bits).
  Usable TCP Payload (MSS) = 9000 − 20 (IPv4) − 20 (TCP) = 8960 Bytes.
  Efficiency η = 8960 ÷ 9038 = 99.14%.

Moving to jumbo frames recovers 4.21% of total network bandwidth that was previously lost to framing tax. More critically, at 100 Gbps line rate, standard 1500-byte frames generate 8.13 million interrupts per second on the host CPU. Jumbo frames reduce this to 1.38 Mpps—slashing CPU interrupt processing overhead by more than 83% and enabling single CPU cores to saturate line rate.

4. Tunnel Encapsulation Taxes & Path MTU Discovery (PMTUD) Blackholes

Modern software-defined data centers (SDDC) heavily employ network virtualization overlays such as VXLAN (RFC 7348), GENEVE, and IPsec ESP. These overlays wrap the entire original packet inside an additional outer transport header:

• VXLAN Overhead (50 Bytes): 14B Outer Ethernet + 20B Outer IPv4 + 8B Outer UDP + 8B VXLAN Header.
• GRE Overhead (24 Bytes): 20B Outer IPv4 + 4B GRE Header.
• IPsec ESP Tunnel Mode (56–72 Bytes): Outer IP header (20B), ESP header (8B), Initialization Vector (8B), ESP trailer and ICV authentication tag (16B), plus encryption cipher block padding.

If the underlying physical underlay network MTU remains clamped at 1500 Bytes, adding a 50-byte VXLAN tunnel forces the tenant payload MTU down to 1450 Bytes, reducing the usable TCP MSS to 1410 Bytes.

When host operating systems assert the Don't Fragment (DF) flag in the IP header, intermediate routers unable to forward frames exceeding the tunnel MTU must drop the packet and return an ICMP Type 3, Code 4 (“Fragmentation Needed and DF set”) message containing the next-hop MTU. If network firewalls block all ICMP traffic, the sending host never receives this signal. Small TCP SYN packets (typically 60 Bytes) successfully negotiate the initial three-way handshake, but large data packets exceeding 1450 Bytes are dropped silently. The TCP session hangs indefinitely—a notorious condition known as a Path MTU Discovery (PMTUD) Blackhole. Resolving this requires either configuring ip tcp adjust-mss clamping on transit firewalls or increasing the physical underlay fabric MTU to 1600+ Bytes (Baby Jumbo).