TCP & UDP — Part 3: Deep Comparison & Advanced Internals

What is this?

We've met both protocols on their own terms: TCP, the careful phone call that confirms everything; UDP, the shout across the room that confirms nothing. Now we put them side by side and ask the questions a real engineer faces — why is TCP slow to start, what exactly is head-of-line blocking, how does QUIC fix it, and which one should carry your traffic.

The one-line idea: TCP bakes one fixed reliability policy into the kernel and applies it to everyone; UDP hands you a blank transport so you (or QUIC) can build precisely the policy your workload needs. The whole comparison flows from that single difference in where the reliability lives.

Keep the analogy running. TCP's reliability is like a phone system where the carrier decides how repeats and pacing work, identically for every call — you can't change it. UDP is like having an open channel and agreeing on your own rules with whoever you're talking to. Sometimes the carrier's defaults are exactly right (a bank transfer). Sometimes you desperately want your own rules (a live video call). This chapter is about telling those situations apart and understanding the machinery underneath.

Both protocols still sit at Layer 4 of the OSI model, under HTTP and friends — but as we'll see, HTTP/3's jump to QUIC-over-UDP is the most consequential transport shift in years.