Reliable or Fast? (TCP vs UDP)

medium

Learn with your AI

Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.

Or teach me the whole lesson:

Why this matters

TCP's TIME_WAIT will pin sockets for 60–120 seconds and silently exhaust ephemeral ports under load — a class of outage that looks like "the database is fine, but new connections fail." UDP gives you raw speed but no delivery guarantees, so anything built on it (DNS, QUIC, video) reinvents reliability at the app layer.

Demo

Inspect TCP connection states and ephemeral port pressure on a running cloud VM.

Try it yourself

Open 200 short HTTP connections with ab -n 200 -c 1 http://example.com/ then count TIME_WAIT sockets — explain why it spikes.
Read net.ipv4.tcp_tw_reuse — flip it from 0 to 1 in a sandbox and document one risk of doing so in production.
Compare the wire size of a DNS query over UDP vs TCP for the same domain using dig +stats — note the difference.

References

# SSH to an EC2 instance, then inspect TCP states
ssh ec2-user@<public-ip>

# Connection state distribution
ss -tan | awk 'NR>1 {print $1}' | sort | uniq -c

# Count sockets stuck in TIME_WAIT
ss -tan state time-wait | wc -l

# Ephemeral port range (default 32768-60999 on Linux)
sysctl net.ipv4.ip_local_port_range

# Test UDP: send a DNS query without a connection
dig @1.1.1.1 +notcp example.com