---

Metadata Card

Attribute	Content
Volume	Vol 4 — Computer Networking
Chapter	Chapter 14: Network Performance & Optimization
Prerequisites	Chapter 1 (Layering), Chapter 3 (TCP), Chapter 7 (DNS/CDN/LB), Chapter 10 (Congestion Control), Chapter 13 (IPv6)
Next	— (Last chapter of Vol 4)
Theory Depth	(2/5) — Tools and practice, no derivations required
Python Relevance	≈40%; focus on Bash toolchain (curl/iperf/ss/tcpdump)
Core Skills	Find root cause of "slowness", TCP kernel tuning, HTTP performance optimization
Code	~100 lines Python + extensive Shell commands

---

Your Progress

"You've traveled from Chapter 1 to Chapter 13 — layering models, Sockets, TCP, HTTP, TLS, DNS, CDN, data link, IP, congestion control, QUIC, debugging, IPv6. Now you string them all together — from a mage saying 'slow' to finding root cause to optimization solutions. This is the practical culmination of your post road knowledge."

Full-Stack Analysis — Where is "Slow" Actually Slow?

Break down request time into five phases:

          DNS      TCP      TLS      TTFB         Content
     ──────┬────────┬────────┬────────┬──────────────┬─────→
Time:    t1       t2       t3       t4            t5

Use curl -w to measure each phase precisely:

Phase	Normal	If Slow	Possible Cause
DNS	< 50ms	> 500ms	Slow DNS server, long recursive chain, no local cache
TCP	< 100ms (same region)	> 500ms	Physical distance, added device latency, loss causing SYN retransmit
TLS	< 100ms (TLS 1.3)	> 500ms	TLS 1.2 full (2-RTT), long cert chain, slow OCSP
TTFB	< 200ms	> 1s	Server-side processing slow (DB query, template rendering), edge cache miss
Content Transfer	Depends on bandwidth	Slow	Bandwidth limit, small congestion window, high packet loss

DNS Optimization

• Use local DNS cache (dnsmasq, systemd-resolved)
• Pre-fetch with <link rel="dns-prefetch"> in HTML
• Use fast DNS resolvers (1.1.1.1, 8.8.8.8)
• CDN Anycast DNS automatically routes queries to nearest node

TCP Connection Optimization

• Keep-Alive: Reuse connections instead of opening new ones
• TCP Fast Open (TFO): Carry data in SYN packet, save 1 RTT
• Reduce physical distance: CDN moves content closer to users

TLS Handshake Optimization

1. Upgrade to TLS 1.3 — reduces from 2 RTT to 1 RTT
2. Enable Session Resumption — cache Session Tickets, 0-RTT on reconnect
3. OCSP Stapling — server provides cert status, no need for client to query OCSP
4. Optimize cert chain — reduce number of intermediate certificates
5. Trim cipher suites — keep only the fastest (AES-GCM, ChaCha20-Poly1305)

TTFB — The Server-Side Performance Black Hole

When TTFB >> DNS + TCP + TLS combined, the problem isn't in the network layer. Stop tuning TCP parameters and check the database.

Cause	Characteristic	How to Check
Slow DB query	TTFB correlates with query count	Enable slow query log
Upstream API slow	Server waiting for others' network	Add tracing on server
Template rendering slow	Large page, complex template	Optimize SSR or add CDN caching
Disk I/O	File reads blocking	iostat / strace
No caching	Every request recalculated	Add cache layer (Redis, Memcached, CDN)

TCP Kernel Tuning

BDP (Bandwidth-Delay Product):

BDP = Bandwidth × RTT

Example: Singapore → Virginia: RTT ≈ 200ms, bandwidth 1Gbps
BDP = (1 × 10⁹ bps) × 0.2s = 200,000,000 bits ≈ 24 MB

If TCP receive window < BDP, bandwidth utilization is window-limited.

# High BDP optimization settings
cat >> /etc/sysctl.d/99-network-performance.conf << 'EOF'
# Receive buffer: min=64KB, default=256KB, max=16MB
net.ipv4.tcp_rmem = 65536 262144 16777216
# Send buffer: min=64KB, default=256KB, max=16MB
net.ipv4.tcp_wmem = 65536 262144 16777216
# Enable window scaling
net.ipv4.tcp_window_scaling = 1
# Disable slow start after idle
net.ipv4.tcp_slow_start_after_idle = 0
# Enable SACK and FACK
net.ipv4.tcp_sack = 1
net.ipv4.tcp_fack = 1
# Increase connection queue
net.core.somaxconn = 65535
EOF
sysctl --system

Measuring with iperf3

# Server side
iperf3 -s

# Client side
iperf3 -c server-ip -t 30 -P 4

---

Traveler's Notes

Between knowing "TCP congestion control uses CUBIC cubic function" and actually finding the root cause when a mage says "slow", there's a gap of ten post roads. This chapter is that road. We don't learn any "new transmission spells" — we learn how to use all the previous knowledge as a troubleshooting toolkit. When you face a network problem, systematically check: DNS → TCP → TLS → TTFB → Content Transfer. Each phase has its own tools and optimization strategies.