FoundationNetworkingFree Preview

How DNS Works — Domain Name System & Global Routing Explained

DNS converts names to addresses and can steer users to healthy, nearby regions.

DNS is the internet's phone book. When you type 'google.com', your browser doesn't know what IP address that is — it knows a name. DNS is the system that translates that name to an IP address so your browser knows where to send the request.

The lookup chain: your device checks its local DNS cache → if miss, asks a recursive resolver (e.g. 8.8.8.8) → the resolver asks a root nameserver ('who handles .com?') → asks a TLD nameserver ('who handles google.com?') → asks Google's authoritative nameserver → finally gets the IP. This chain sounds slow but is typically fast because resolvers cache responses aggressively using TTL (Time To Live).

For system design, DNS is more interesting than a simple lookup: it's a global routing layer. GeoDNS returns different IPs based on where the user is — European users get a European data center, US users get a US one. DNS failover can route away from unhealthy regions. Anycast assigns the same IP to servers in multiple data centers, so the network's routing protocol automatically sends users to the nearest one (how Cloudflare works).

Key concepts

recursive resolverrootTLDauthoritativeTTLAnycastGeoDNS

Step-by-step approach

1
Client sends a DNS query to the OS stub resolver, which checks the local cache first (using TTL from the previous lookup).
2
On cache miss, the stub resolver forwards to the configured recursive resolver (ISP, 8.8.8.8, or corporate).
3
Recursive resolver walks the DNS hierarchy: root nameserver → TLD nameserver → authoritative nameserver.
4
Authoritative nameserver returns the IP address (A record) and a TTL. The recursive resolver caches the response for TTL seconds.
5
For multi-region deployments, configure GeoDNS to return different A records based on the client's geographic location, or use Anycast for automatic nearest-server routing.

Key trade-offs

Short TTL vs. long TTL

Short TTL (60s) enables fast failover — DNS propagates quickly when a server goes down. Long TTL (3600s) reduces resolver load and speeds up lookups for users. Design around your failover requirements.

GeoDNS vs. Anycast

GeoDNS is application-level routing (you control what IP each region gets). Anycast is network-level routing (BGP automatically routes to the nearest server sharing the same IP). Anycast is harder to set up but more resilient.

DNS-based load balancing vs. L7 load balancing

DNS round-robin is simple but doesn't health-check. L7 load balancers health-check and route only to healthy backends. Use both layers for production reliability.

Common pitfalls

⚠

Not accounting for DNS caching in failover: even with a 60s TTL, some resolvers and clients ignore TTL and cache longer. Plan for 5-15 minutes of propagation time.

⚠

Using DNS as the only layer of load balancing: DNS returns IPs equally but doesn't know which backend is under load. Combine with a real load balancer.

⚠

Forgetting negative TTL: a failed DNS lookup is also cached (for the negative TTL period), causing repeated failures even after you fix the issue.

Interview questions on this topic

How would you use DNS to route users to the nearest data center for a globally distributed service?

Explain how you'd design a failover strategy that uses DNS to redirect traffic away from a failing region.

Practice answering these with AI feedback → Start on CrackLab

Continue learning — explore 207 more topics on CrackLab DesignHub

From Classic HLD designs (Twitter, YouTube, Uber) to LLD patterns, distributed systems, databases, and company case studies.

Start Learning Free →See all 208 topics