Object tracking in distributed systems

Object mobility (or process migration) is a very important feature in modern distributed systems. By allowing objects to migrate from one machine to another preemptively, the sys-tem can provide several benefits. These include sharing load, reducing communication cost, increasing availability, and improving invocation performance. One problem with object mobility is object tracking. Several schemes exist to handle the problem. A forwarding address scheme is used in DEMOS/MP and Emerald. To reduce the number of forwarding messages, DEMOS/MP uses an urgent updating policy to compress the message path. Whenever an object is moved, an updating message is sent to the last node immediately. In Emerald, a lazy updating policy is adopted. When an object migrates, no updating message is sent. The new location is piggybacked on each reply message. This can also compress the message path for later invocations. The difference between the two is that DEMOS/MP places the cost on object migration while Emerald places the cost on object invocation. They both use the same updating policy for all objects. We adopt a philosophy in which objects with different behaviors use different updating policies. Objects are divided into two groups: active objects and quiet objects. Active objects are defined as objects which move more often than they are invoked. Quiet objects are invoked more often than they are moved. To optimize object moving and invoking, active objects should use a lazy updating policy while quiet objects should use an urgent updating policy. We call this policy the adaptive updating policy. The object tracking function is separated from Emerald and implemented as an independent protocol layer, OFP. A reliable datagram protocol, RDP has been designed to support OFP. Experiments are done to validate the protocols' performance. The protocols are implemented on the x-kernel platform.Science, Faculty ofComputer Science, Department ofGraduat