HTTP Keep-Alive, TCP Keep-Alive

This is a feature that allows a single TCP connection to be reused for multiple HTTP requests at the HTTP layer.

In other words, it's a performance optimization that prevents repeating the TCP 3-way handshake for every request.

The client includes Connection: keep-alive in the HTTP request header.
The server maintains the connection for a certain period.
Multiple HTTP requests/responses are exchanged over the same TCP connection. The server closes the connection after an Idle Timeout (e.g., 60s).

Its purpose is to reduce RTT, improve latency, and enhance performance through connection reuse between the server and client.

This is a network-level feature to check if a TCP socket is dead (i.e., if the peer has disappeared).

In other words, its purpose is for the system to check if that connection is still alive.

The kernel sends a keepalive probe packet after a certain period.
If there's no response from the peer, it retransmits.
If it fails a specified number of times, the kernel forcibly closes the socket (ETIMEDOUT).

Transport TCP feature
Very long default period
- tcp_keepalive_time (default 7200 seconds, 2 hours)
- tcp_keepalive_intvl
- tcp_keepalive_probes

Item	HTTP Keep-Alive	TCP Keep-Alive
Layer	Application(HTTP)	Transport(TCP)
Purpose	Connection reuse (performance improvement)	Check connection liveness
Timing	Application-configured Idle Timeout	OS kernel configuration-based (minutes to hours)
Packets Used	HTTP requests/responses	TCP keepalive probe
When Disconnected	Server attempts Idle Timeout	Kernel closes socket on probe failure
Prevents NAT/Firewall idle drop	Impossible	Possible

HTTP Keep-Alive = Reuse
- Maintains the connection for continuous use.
- A feature for performance.
- No role in detecting if the connection is dead.
TCP Keep-Alive = Liveness Check
- Checks if the peer is alive.
- Detects network disconnections and server crashes.
- Managed by the kernel.