Web Acceleration and Protocol Optimization: Practices and Future Directions

The talk from the 2017 ArchSummit introduced common web performance problems—large page size, excessive elements, and network latency—and highlighted that many issues stem from protocol inefficiencies.

It examined HTTP/1.1’s single‑connection serialization and header redundancy, then contrasted these with HTTP/2’s multiplexing, header compression (HPACK), and server push, noting the performance gains when using HTTPS with HTTP/2.

TLS was identified as a major latency source due to its multi‑RTT handshake; the presentation covered session‑id and session‑ticket mechanisms, false‑start, and OCSP stapling, as well as TLS 1.3’s 0‑RTT and 1‑RTT handshakes.

Transport‑layer optimizations such as TCP Fast Open, increasing the initial congestion window, and dynamic record sizing were described, along with their limited deployment constraints.

Practical recommendations included using a single connection, minimizing domain names, leveraging server push instead of inlining, configuring TLS 1.2 cipher suites for HTTP/2, and employing pre‑connect techniques (link tags or JavaScript) to hide connection setup latency.

The speaker concluded that while HTTPS with HTTP/2 already outperforms HTTP/1.1, the next performance frontier lies in QUIC, which combines UDP transport with HTTP/2 features and promises reduced handshake overhead and mitigated head‑of‑line blocking.