Home Load Balancing: L4/L7, HAProxy, NGINX, AWS ALB/NLB, GSLB และ Application Delivery Load Balancing: L4/L7, HAProxy, NGINX, AWS ALB/NLB, GSLB และ Application Delivery
Load Balancing: L4/L7, HAProxy, NGINX, AWS ALB/NLB, GSLB และ Application Delivery
Load Balancing กระจาย traffic ไปยัง backend servers เพื่อ scalability และ high availability L4 Load Balancing ทำงานที่ transport layer (TCP/UDP), L7 ทำงานที่ application layer (HTTP/HTTPS), HAProxy เป็น open-source high-performance load balancer, NGINX เป็น web server + reverse proxy + LB, AWS ALB/NLB เป็น managed cloud load balancers, GSLB กระจาย traffic ข้าม data centers และ Application Delivery รวมทุกอย่างเป็น platform
ทุก web application ที่มี traffic สูง ต้องใช้ load balancer : single server handle ได้จำกัด (เช่น 10,000 concurrent connections), load balancer กระจาย traffic ไปหลาย servers → scale horizontally ถ้า server ตัวหนึ่ง fail → load balancer route ไป servers ที่เหลือ → zero downtime Netflix ใช้ load balancing หลายชั้นเพื่อ serve 250M+ subscribers
L4 vs L7 Load Balancing
Feature
L4 (Transport)
L7 (Application)
Layer
TCP/UDP (IP + Port)
HTTP/HTTPS (URL, headers, cookies)
Decision Based On
Source/Dest IP, Port
URL path, hostname, headers, cookies, content type
Performance
Faster (no payload inspection)
Slower (parse HTTP, SSL termination)
SSL Termination
Pass-through (no inspection)
Terminate SSL → inspect → re-encrypt or plaintext to backend
Content Routing
ไม่ได้ (ไม่เห็น content)
/api → API servers, /static → CDN, /ws → WebSocket servers
Use Case
TCP services (database, SMTP, gaming)
Web applications (HTTP routing, API gateway)
Examples
AWS NLB, HAProxy TCP mode, LVS
AWS ALB, HAProxy HTTP mode, NGINX, Envoy
Load Balancing Algorithms
Algorithm
How
Best For
Round Robin
Rotate ทีละ server ตามลำดับ
Equal-capacity servers, stateless apps
Weighted Round Robin
Round robin แต่ server แข็งกว่าได้ weight มากกว่า
Mixed-capacity servers
Least Connections
ส่งไป server ที่มี active connections น้อยที่สุด
Long-lived connections (WebSocket, database)
IP Hash
Hash source IP → always same server (sticky)
Session persistence without cookies
Least Response Time
ส่งไป server ที่ response เร็วที่สุด
Heterogeneous backends (different response times)
Random
Random selection (with power of two choices)
Large server pools (simple, effective)
HAProxy
Feature
รายละเอียด
คืออะไร
Open-source, high-performance TCP/HTTP load balancer and proxy
Performance
Handle 1M+ concurrent connections, 100K+ requests/sec on single instance
Modes
TCP mode (L4) + HTTP mode (L7) — both in single config
Health Checks
TCP, HTTP (status code, body), custom scripts — remove unhealthy backends
SSL
SSL termination, SSL passthrough, SSL re-encryption
ACLs
Rich ACL system for routing: path, headers, cookies, source IP, SNI
Stats
Built-in stats page (real-time dashboard) + Prometheus exporter
Used By
GitHub, Stack Overflow, Reddit, Airbnb, Instagram
NGINX
Feature
รายละเอียด
คืออะไร
Web server + reverse proxy + load balancer + HTTP cache
Architecture
Event-driven, non-blocking I/O → handle 10,000+ concurrent connections efficiently
Load Balancing
Round robin, least connections, IP hash, random (upstream module)
L7 Features
URL routing, header manipulation, caching, rate limiting, gzip compression
NGINX Plus
Commercial version: active health checks, session persistence, DNS service discovery, dashboard
vs HAProxy
NGINX: web server + LB (better for static content, caching) | HAProxy: pure LB (better for complex routing)
Market Share
34%+ of all websites use NGINX (most popular web server)
AWS Load Balancers
Type
Layer
Use Case
Features
ALB (Application)
L7 (HTTP/HTTPS)
Web apps, microservices, containers
Path/host routing, WebSocket, gRPC, WAF integration
NLB (Network)
L4 (TCP/UDP/TLS)
High performance, static IP, gaming
Millions of RPS, static IP/EIP, ultra-low latency
CLB (Classic)
L4 + L7 (legacy)
Legacy EC2-Classic
Deprecated — migrate to ALB/NLB
GWLB (Gateway)
L3 (IP)
Inline security appliances
Route traffic through firewalls/IDS transparently
GSLB (Global Server Load Balancing)
Feature
รายละเอียด
คืออะไร
Distribute traffic across multiple data centers/regions globally
DNS-Based
Return different IP based on client location, health, load → DNS resolves to nearest DC
Anycast
Same IP advertised from multiple locations → BGP routes to nearest (CDN approach)
Policies
Geographic (nearest), weighted (ratio), failover (primary/backup), latency-based
Health Checks
Monitor each DC’s health → remove unhealthy DC from DNS/anycast
Products
AWS Route 53, Cloudflare, F5 BIG-IP DNS, NS1, Akamai GTM
ทิ้งท้าย: Load Balancing = Scale, Availability, Performance
Load Balancing L4 vs L7: L4 (TCP/UDP, fast, no content inspection) vs L7 (HTTP, content routing, SSL termination) Algorithms: round robin, weighted, least connections, IP hash, least response time, random HAProxy: pure LB (1M+ connections), rich ACLs, TCP+HTTP modes, used by GitHub/Reddit NGINX: web server + LB + cache, event-driven, 34% market share, NGINX Plus for enterprise AWS: ALB (L7 web), NLB (L4 performance), GWLB (security appliances) GSLB: DNS-based or anycast, geographic/latency routing, multi-DC failover Key: L4 for raw performance, L7 for intelligent routing — most apps need L7 with proper health checks and algorithms
อ่านเพิ่มเติมเกี่ยวกับ Kubernetes Networking CNI Service Mesh Ingress และ Cloud Networking VPC Transit Gateway ที่ siamlancard.com หรือจาก icafeforex.com และ siam2r.com
