Global server load balancing (GSLB) in HLD - Scalability & System Analysis

Scalability Analysis - Global server load balancing (GSLB)

Growth Table: Global Server Load Balancing (GSLB)

Users	Traffic Characteristics	GSLB Impact	Infrastructure Changes
100 users	Low requests, regional access	Simple DNS-based routing sufficient	Single data center, basic load balancer
10,000 users	Moderate requests, some geo diversity	Need multiple regional data centers, basic GSLB	Deploy multiple data centers, DNS with geo-location
1,000,000 users	High requests, global access, peak traffic	GSLB must handle failover, latency optimization	Multiple global data centers, health checks, latency-based routing
100,000,000 users	Very high requests, global with spikes	GSLB must scale massively, integrate CDN, DDoS protection	Global distributed data centers, multi-layer load balancing, advanced monitoring

First Bottleneck

The first bottleneck in GSLB systems is usually the DNS resolution and health check system.

At scale, DNS servers can be overwhelmed by queries, and stale or slow health checks can cause poor routing decisions.

Also, network latency and inconsistent health data can cause traffic to route to unhealthy or distant servers, increasing user latency.

Scaling Solutions

• DNS Scaling: Use Anycast DNS and multiple authoritative DNS servers worldwide to handle query load.
• Health Checks: Implement distributed, frequent, and fast health checks with caching to reduce load and improve accuracy.
• Latency-based Routing: Use real-time latency measurements and geo-IP databases to route users to the closest healthy server.
• Failover: Automatic failover to backup data centers when primary ones are down.
• Integration with CDN: Offload static content to CDN to reduce load on origin servers and improve global performance.
• Load Balancer Hierarchy: Combine global load balancers (GSLB) with local load balancers for efficient traffic distribution.
• DDoS Protection: Deploy network-level protections to handle large traffic spikes and attacks.

Back-of-Envelope Cost Analysis

• Requests per second (RPS): At 1M users, assuming 1 request per second per 10 users, ~100,000 RPS globally.
• DNS Queries: Each user may generate multiple DNS queries; DNS servers must handle millions of queries per second at large scale.
• Bandwidth: For 100M users, assuming 100 KB per request, bandwidth needed is ~1 TB/s globally, requiring multi-region data centers and CDNs.
• Storage: Mainly for logs and health data; scalable with cloud storage solutions.
• Infrastructure: Multiple global data centers, Anycast DNS, CDN subscriptions, monitoring and security services.

Interview Tip

Start by explaining the user growth and traffic patterns globally.

Identify the first bottleneck (DNS and health checks) and explain why.

Discuss scaling solutions step-by-step: DNS scaling, health checks, latency routing, failover, CDN integration.

Use real numbers to justify your choices and show understanding of global infrastructure challenges.

Self Check Question

Your DNS servers handle 1000 queries per second. Traffic grows 10x. What do you do first?

Answer: Deploy additional authoritative DNS servers with Anycast IPs to distribute query load and reduce latency. Also, implement DNS caching and optimize TTL values to reduce query frequency.

Key Result

GSLB scales by distributing DNS and traffic globally, but DNS query load and health check accuracy are the first bottlenecks; scaling requires Anycast DNS, distributed health checks, latency-based routing, and CDN integration.