Fragments

A Quick Comparison of DDC Site-Selection Methods

August 12th, 2008 | by Mitja Robas |

When setting up a distributed data center (DDC), the architect has to decide which method will be used for site selection to distribute client traffic between the sites. The distribution is done using global server load balancing (GSLB), typically via HTTP redirect, DNS-based site selection or route health injection (RHI) combined with L3 routing.

What are the advantages of each method?
HTTP redirect:

• independent of any GSLB devices or mechanisms
• offers persistence to the selected site by default
• can be complemented with any other method for more sophisticated site selection

DNS-based site selection:

• works with any application (FTP, web, mail, etc.)
• requires minimal configuration changes in the DNS authoritative server
• can take load or data center size into account
• offers multiple load-balancing methods (round robin, weighted round robin, site proximity, source address etc.)

RHI with L3 routing:

• does not reply on DNS, and thus also supports legacy applications
• has good reconvergence, especially if the interior gateway protocol is fine-tuned
• independent of the application protocols
• addresses also the client’s browser stale IP addresses

What are the disadvantages of each method?
HTTP redirect:

• relies on HTTP and is thus protocol-specific
• requires additional fully qualified domain names (FQDNs) for redirection
• is not »transparent« to the user, who might bookmark a specific location and thus lose automatic failover

DNS-based site selection:

• client DNS caching can break the desired site-selection operation
• guaranteed session persistency is harder to implement
• requires a »site selector« device

RHI with L3 routing:

• relies on host routes (32 bits) that cannot be propagated through the Internet
• application-aware devices such as load balancers must be tightly integrated with routers
• lacks intelligent load-balancing functionality (more suitable for disaster recovery)