Load-Balancing Method


As one NAS machine reaches its optimal handling capacity, subsequent requests are sent to another NAS machine with duplicate application components. With multiple servers handling requests, response time increases.

Response Time

Response-time load balancing bases decisions about where to send requests on how fast a server responds.

  • Initially, response time requests are handled round-robin style.
  • After NAS gathers preliminary response-time statistics for each of the servers, it begins using response time to route requests.
  • NAS maintains the response-time statistics in a sliding window, thus compensating for spikes or other timing anomalies.

Per Component

The Per Component value maintains response-time statistics for each application component.
 

Per Server

The Per Server value calculates response time for the entire server.

 

User Defined

When a Netscape Application Server evaluates itself, it then broadcasts the information. This broadcast is sent to a multicast IP address and port number held in the registry.

By listening for the broadcasts of other NAS and considering its own information, a NAS creates a load-balancing table. This table is under constant reconstruction as information is updated and processed.

Server Load

The hardware resources of the machine that houses the Netscape Application Server are evaluated. The four components evaluated are CPU, memory, Disk I/O, and number of requests in the queue.
 

Application Component Criteria

There are four criteria for Application Component performance:

  • Cached results
  • Available lowest average execution time
  • Most recently executed
  • Fewest executions