| 摘要 |
<p>A load balancing architecture routes object creation requests to server computers based on the object's class or some other characteristic. The architecture operates transparently to the object and the client program creating the object. A monitoring service at the servers observes a processing metric such as response time and provides the observations to a router computer. A load balancing engine at the router computer analyzes the observations to dynamically adjust routing, directing object creation requests to server computers having favorable performance. Analysis of observations is delayed to facilitate blending and dampening. Although a default load balancing engine is provided, software developers can specify a different engine per object class, and the engines can be tailored to the behavior of a particular object class. Load balancing is extensible to threads and processes. If performance at a server falls below an acceptable level, a rebalance policy transparently sends a rebalance message, breaking the connectionto the server.</p> |
