QUIC

QUIC is a transport protocol, originally from Google and now an IETF standard, that provides TCP-like reliability and ordering but is built on UDP instead of being a peer to it. The motivating problem is head-of-line blocking: over a single TCP connection carrying multiple logical streams (as HTTP/2 does), one lost packet stalls all of them until it's retransmitted, because TCP only knows about one ordered byte stream, not the independent streams multiplexed inside it. QUIC gives each stream its own sequencing, so a lost packet on stream A never blocks stream B from being delivered to the application.

QUIC also folds the transport and TLS handshakes into one round trip instead of two sequential ones (TCP's three-way handshake, then a separate TLS negotiation on top), and it can often resume a previous connection with zero additional round trips at all — meaningful savings on high-latency or lossy mobile networks where every round trip counts. Because QUIC runs in userspace over UDP rather than in the kernel like TCP, it can also evolve faster: shipping a new congestion-control algorithm is a library update, not an OS update.

It underlies HTTP/3, which is effectively "HTTP semantics over QUIC" the way HTTP/2 was "HTTP semantics over TCP with multiplexing." Connection migration is one of its more distinctive features: a QUIC connection is identified by a connection ID rather than by the traditional IP-and-port tuple, so a phone can switch from Wi-Fi to cellular mid-download without dropping the connection, something TCP has no clean way to express.

The cost is CPU: encrypting and processing every packet in userspace is more work than the kernel's optimized TCP stack, which is part of why adoption took years even after the spec stabilized.