What is TCP window scaling and large receive offload?
Why Interviewers Ask This
Interviewers ask this to evaluate whether you have the depth of knowledge needed to mentor others and lead technical decisions. The expected answer goes beyond definitions into practical implications and real-world consequences.
Answer
The original TCP header uses a 16-bit receive window field — maximum 65,535 bytes. On high-bandwidth, high-latency links (high BDP — Bandwidth-Delay Product), this limits throughput severely. Example: 100 Gbps link, 100ms RTT — optimal window is 100 Gbps × 0.1s = 10 GB, far exceeding the 64 KB limit. TCP Window Scaling (RFC 1323) adds a scaling factor (0-14) negotiated during the 3-way handshake, allowing windows up to 1 GB (65535 × 2^14). Enabled by default in modern OS. Large Receive Offload (LRO) and Generic Receive Offload (GRO) are NIC and OS features that aggregate multiple incoming TCP segments into a single larger packet before passing to the kernel — reducing CPU overhead for high-throughput workloads. TCP Segmentation Offload (TSO) allows the OS to send large segments; the NIC splits them into MTU-sized pieces — reducing CPU cycles. These optimizations are critical for achieving 10/40/100 Gbps network throughput on servers without saturating CPUs.
Pro Tip
Demonstrate both theoretical understanding and practical experience. Say what it is, then give an example of how you actually used it in a Networking codebase.