HTTP Keep-Alive vs TCP Keepalive: Key Differences Explained

Hello, I am Xiao Lin.

A reader asked two interview questions: "How does MySQL store NULL values?" and "What is the difference between HTTP Keep-Alive and TCP Keepalive?" The first concerns MySQL record storage; the second concerns persistent connections.

The two concepts are completely different and operate at different layers:

HTTP Keep-Alive is implemented at the application layer (user space) and is called an HTTP persistent connection.

TCP Keepalive is implemented at the TCP layer (kernel space) and is called the TCP keepalive mechanism.

HTTP Keep-Alive

HTTP follows a request‑response model. Each request normally creates a TCP connection, sends the request, receives the response, and then closes the connection, which is called an HTTP short connection.

To avoid the overhead of repeatedly establishing TCP connections, HTTP Keep-Alive allows multiple requests and responses to share a single TCP connection, forming an HTTP long connection.

In HTTP/1.0, Keep-Alive must be enabled by adding the header:

<code>Connection: Keep-Alive</code>

The server also replies with the same header to keep the connection alive.

<code>Connection: Keep-Alive</code>

From HTTP/1.1 onward, Keep-Alive is enabled by default; to disable it, send:

<code>Connection: close</code>

HTTP long connections reduce TCP connection overhead and enable HTTP pipelining, where a client can send multiple requests without waiting for each response, improving overall response time. However, the server still responds in order, which can cause head‑of‑line blocking.

To avoid wasting resources when a connection remains idle, web servers provide a

keepalive_timeout

parameter that closes the connection after a period of inactivity.

TCP Keepalive

TCP Keepalive is the kernel‑level keepalive mechanism. When no data is exchanged for a certain period, the TCP stack sends probe packets to the peer.

If the peer is alive, it responds, and the keepalive timer resets.

If the peer does not respond after several probes, TCP marks the connection as dead.

Applications must enable TCP keepalive by setting the

SO_KEEPALIVE

socket option.

Summary

HTTP Keep-Alive (HTTP long connection) is implemented by the application and allows multiple HTTP requests/responses over a single TCP connection, reducing the overhead of short connections.

TCP Keepalive is implemented by the kernel and periodically probes idle connections to determine if they are still alive, closing them when they are not.