Using Layered Architectures to Create Highly Available Infrastructures

{include:DSEE Banner}
{toc:type=flat|separator=bracket}

Use load balancers and [Sun Java System Directory Proxy Server] to create high available infrastructures. Adding isolation between layers of servers increases the chances of components detecting failures in components below them. This allows administrators to create infrastructures that can be modified without knowledge to service clients.

h1. Scenarios

h2. Scenario One: Clients Connect Directly To Directory Server

!l1.png|width=640!

When clients connect directly to Directory Server:

* Clients have knowledge of directory services infrastructure, IP addresses and hostnames
* No encapsulation of service infrastructure
* Client must provide own failure detection and failover mechanisms - increasing client complexity
* Uncommanded service termination causes an outage
* Clients must provide own load balancing - increasing client complexity
* No virtualization capability
* Unpredictable service uptime results in unclear service level objectives
* Simplistic solution suitable for non-critical environments
* Replication latency and loose consistency play major roles
* Commanded outages require at least notification of application server operators, possibly configuration change because applications have knowledge of the infrastructure
* Might need SSL accelerators on hardware

h2. Scenario Two: Clients Connect to Directory Server via Load Balancers


!l2.png|width=640!

* Better than Scenario One from an availability perspective
* Client has no knowledge of directory services infrastructure, IP addresses and hostnames
* Encapsulation of service infrastructure, easier to change the infrastructure without client knowledge
* Load balancer provides load balancing and service interruption detection
* Uncommanded service termination may not cause an outage - except for persistent connections
* Commanded service termination possible without outage - except for persistent connections
* Increased complexity and cost but higher availability
* Still no virtualization capability
* Slightly predictable service uptime
* Replication latency and loose consistency play major roles
* [SSL] sessions can be terminated at load balancers

h2. Scenario Three: Clients Connect To Directory Servers via Load Balancers and Directory Proxy Server


!l3.png|width=640!

* Highest availability than the previous two Scenarios

* Very complex environment, requires multiple monitoring tools/agents
* Virtualization possible
* More predictable service uptime
* Complex interactions realized via Directory Proxy Server
* Isolated environment increases availability and latency
* In some cases (but not all) replication latency not a factor due to [client affinity].

h1. Contributors

{contributors-summary}