Frontend System Design · Lesson 1 of 86 · 4 min
HTTP/1.1 vs HTTP/2 vs HTTP/3 (QUIC)
This is a fundamental check. If you can't explain why HTTP/2 was created and what HTTP/3 fixes, the interviewer will question your understanding of web performance.
This lesson gives you the core story of HTTP's evolution. You need to be able to explain the performance bottleneck in each version and how the next one solved it. This isn't just history; it's the "why" behind modern frontend optimization.
HTTP/1.1: One request at a time
HTTP/1.1 handles one request and response at a time on a single TCP connection. If you need to download an HTML file, a CSS file, and an image, they are requested sequentially. The next request can't start until the previous response is fully received.
This creates a bottleneck called Head-of-Line Blocking (HOLB). A slow or large response blocks everything behind it on that connection.
To work around this, browsers open multiple TCP connections to the same domain (typically 6). This allows for some parallelism, but each new connection has an overhead cost: the TCP and TLS handshakes, which we'll cover in the next lesson.
HTTP/2: Multiplexing fixes one problem, but not the big one
HTTP/2's main goal was to solve HTTP/1.1's HOLB. Its key feature is multiplexing, which allows multiple requests and responses to be sent and received in parallel over a single TCP connection.
The browser can fire off requests for all the assets on a page at once, and the server can send them back as they become available. A small, quick image doesn't have to wait for a large, slow script. This eliminates the need for multiple connections, saving on handshake overhead.
HTTP/2 also introduced a more efficient binary protocol (instead of text) and header compression (HPACK) to reduce data transfer.
The trap: A lot of developers think HTTP/2 completely solved Head-of-Line Blocking. It did not. It solved HOLB at the application layer (HTTP requests), but the problem still exists at the transport layer (TCP).
Because all these parallel streams are multiplexed over one TCP connection, a single lost TCP packet will cause TCP to halt all streams until that packet is retransmitted. One lost packet for an image can delay the delivery of your CSS and JavaScript.
HTTP/3: Fixing the real head-of-line blocking with QUIC
HTTP/3 fixes the TCP-level HOLB problem by switching the transport layer. Instead of TCP, it uses a new protocol called QUIC, which runs on top of UDP.
QUIC re-implements the features of TCP (like reliability and congestion control) but with a crucial difference: its streams are independent. If a packet for one stream is lost, it only blocks that specific stream. Other streams on the same connection can continue processing data.
This is the definitive solution to HOLB.
Another major benefit of QUIC is a faster connection setup. It combines the TCP and TLS handshakes into a single, faster process. For subsequent connections to a known server, it can even achieve a "zero round-trip time" (0-RTT) handshake, sending application data in the very first packet. This is a significant latency reduction, especially on mobile networks.
How this changes your job (or doesn't)
You don't write code like fetch(url, { version: 'HTTP/3' }). The browser and server negotiate the highest protocol they both support automatically. Your job is to understand the implications and adjust your optimization strategies.
The big anti-pattern to call out in an interview is domain sharding. This was a hack for HTTP/1.1 where you'd spread assets across assets1.example.com, assets2.example.com, etc., to get around the 6-connection limit. With HTTP/2 and HTTP/3, this is counter-productive. It breaks the single-connection multiplexing benefit and adds unnecessary DNS lookups and connection overhead.
Instead, you should leverage resource hints. They work beautifully with modern protocols.
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Optimized Page</title>
<!--
preconnect: Establishes a connection to a domain early.
With HTTP/2 or HTTP/3, this single connection can then be
reused for all subsequent requests to that domain.
-->
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link rel="preconnect" href="https://cdn.example.com">
<!--
preload: Fetches a critical resource with high priority.
HTTP/2 and HTTP/3 can multiplex this request alongside others
without blocking.
-->
<link rel="preload" href="/styles/main.css" as="style">
<link rel="stylesheet" href="/styles/main.css">
<link rel="preload" href="/scripts/app.js" as="script">
<script src="/scripts/app.js" defer></script>
</head>
<body>
<h1>Hello, World!</h1>
</body>
</html>
preconnect pays off more because the connection it establishes can be efficiently reused. preload is more effective because the request can be multiplexed without getting stuck in a queue. You are giving the browser hints that align with how the underlying protocol works best.
Try it
Open DevTools in your browser and go to the Network tab. Right-click the table headers and enable the "Protocol" column. Visit a few major sites (Google, Facebook, Cloudflare, Netflix). See which protocol they use ("h2" for HTTP/2, "h3" for HTTP/3). Find a site using HTTP/3 and compare its perceived loading behavior to an HTTP/2 site, especially on a throttled connection.
Next: We'll break down the setup costs that HTTP/3 reduces in "TCP and TLS Handshakes: Connection Setup Costs".
Lock it in — apply this lesson under interview conditions.