Data Centers in System Design

System Design · Days 31–40 · 3 min read

Video Reference: Data Centers in System Design

1. Asli Problem Kya Hai? (The Core Why)

Socho aap Swiggy ke delivery manager ho. Har roz lakhon orders aate hain, aur log chahte hain ki unka food exactly 30‑45 minutes mein pohonch jaye. Agar aapke servers ek hi data center mein hi ho, to ek time zone ke customers ko serve karte hue aapko latency aur availability ki problem face karni padegi.

Ek aur example: IRCTC ka booking system. Jab peak season (Christmas, Diwali) mein users booking karte hain, to agar sirf ek data center hota, to traffic spikes ke kaaran denial of service ho sakta, aur users ko tickets milne mein hours lag sakte hain.

Yeh problem global reach aur high reliability chahiye, lekin ek single point of failure se downtime aur user frustration badh jata hai. Data Centers ka idea hi ye hai ki geographically distributed servers se latency kam ho, availability badhe, aur fault tolerance mile.

2. Core Architecture & Mechanisms

2.1 Geographic Distribution

Data Centers multiple regions (US East, EU West, Asia‑Pacific) mein spread hote hain.

Latency: User ko nearest DC se response milta hai.
Read Replica: Main DB se data copy karke read load reduce karte hain.

2.2 Load Balancing & Traffic Routing

Global Load Balancer: DNS‑based routing (e.g., Amazon Route 53) jo user ko nearest DC dikhaata hai.
Local Load Balancer: L4/L7 balancer (NGINX, Envoy) jo intra‑DC traffic distribute karta hai.

2.3 Data Replication & Consistency

Multi‑Region Replication: Data ko asynchronously ya synchronously replicate kiya jata hai.
Consistency Models: Eventual consistency (Zerodha’s order book) vs strong consistency (Banking transactions).
Conflict Resolution: Vector clocks, CRDTs, or application‑level merges.

2.4 Failure Handling & Disaster Recovery

Failover: Automatic switchover to standby DC when primary fails.
Backup & Restore: Regular snapshots aur cross‑region backups.
Chaos Engineering: Randomly shutting down instances to test resilience.

2.5 Security & Compliance

Encryption at Rest & Transit: AES‑256, TLS 1.3.
Regulatory Compliance: GDPR for EU, PDP for India.

3. Architecture Flow

graph TD
    A[User Request] -->|DNS| B{Global LB}
    B -->|US East| C[US East DC]
    B -->|EU West| D[EU West DC]
    B -->|APAC| E[APAC DC]
    C -->|Local LB| F[App Servers]
    D -->|Local LB| G[App Servers]
    E -->|Local LB| H[App Servers]
    F -->|DB Read| I[Read Replica]
    G -->|DB Read| J[Read Replica]
    H -->|DB Read| K[Read Replica]
    I -->|Write| L[Primary DB]
    J -->|Write| L
    K -->|Write| L
    L -->|Replication| I
    L -->|Replication| J
    L -->|Replication| K

4. Production Case Study

WhatsApp uses a global data center strategy with multiple regions (US, EU, APAC).

Every message is stored in a primary DC, then replicated to secondary DCs using gossip protocol.
They employ read replicas for status checks and global load balancers to route users to the nearest DC, reducing latency to < 100 ms.
In case of a DC outage, traffic automatically reroutes to the next best DC, ensuring 99.999% uptime.

5. Trade-offs (Pros vs Cons)

Pros	Cons
Low Latency – users hit nearest DC.	Complexity – multi‑region architecture needs careful design.
High Availability – failover across regions.	Higher Cost – multiple data centers, replication bandwidth.
Disaster Recovery – cross‑region backups.	Consistency Challenges – eventual consistency may affect business logic.
Scalability – add more DCs as traffic grows.	Data Sovereignty – regulatory constraints across countries.
Load Distribution – balanced traffic reduces hotspots.	Operational Overhead – monitoring, logging across regions.

6. System Design Interview Cheat Sheet

Geographic Spread: Explain why multiple data centers help with latency and availability.
Load Balancing: Distinguish between global DNS LB and local L4/L7 LB.
Replication Models: Eventual vs strong consistency; mention vector clocks or CRDTs.
Failover Mechanisms: Automatic switchover, backup strategies, and chaos testing.
Cost vs Reliability Trade‑off: Be ready to discuss how to balance budget with SLAs.