Global Server Load Balancing (GSLB)
GSLB answers DNS queries with the best site for a client, using site health, metrics, and persistence. The exam tests the relationship between GSLB sites, services, and virtual servers, plus what happens when a site fails.
Core concepts
- A GSLB site has a site IP. Local and remote sites exchange metrics through MEP. A GSLB virtual server is bound to a domain and contains GSLB services that point to local VIPs or remote site services.
- Methods include active-active options such as round robin, least connection, least response time, and static proximity, plus active-passive designs. Fallback and spillover handle capacity and failure.
- Connection proxy lets a client connect to the local site IP, which proxies the connection to a remote site. This helps when DNS TTL is long or the client cannot reach the chosen site directly.
- Site persistence, using a cookie or source IP, keeps a user on one site across requests.
- An ADNS service listens for DNS. Delegation from the parent zone uses NS records or a CNAME.
Worked example
Two data centers run active-active GSLB with a 30-second DNS TTL. Site A fails. MEP marks the site down, and the GSLB virtual server begins returning site B. Clients that cached site A wait up to the TTL before retrying. With connection proxy, the DNS answer can be the local site IP and the proxy redirects to the available remote site, reducing exposure to client-side DNS caching. The reasoning is that GSLB is DNS-based, so TTL and resolver caching are part of the failover time, not just MEP detection.
Traps that cost marks
Forgetting the ADNS service means no DNS answers. MEP reachability between site IPs is required for health exchange. A GSLB service is not the same as a local load-balancing service. Route health injection applies to local VIP advertisement, not to remote site selection by itself. GSLB virtual servers must be bound to the correct domain. Backup virtual servers and fallback methods serve different purposes. Spillover thresholds can send users to a remote site before the local site is down.