Systems management means the management of heterogeneous subsystems of network devices, processing platforms, distributed applications, and other components found in communications and computing environments. Current system management relies on presenting a model to the user of the managed system which should accurately reflect the current state of the system and should ideally be capable of predicting the future health of the system. System management relies on a combination of asynchronously generated alerts and polling to determine the health of a system .
The management application presents state information such as link state, buffer fill and packet loss to the user in the form of a model . The model can be as simple as a passive display of nodes on a screen or a more active model which allows displayed nodes to change color based on state changes, or react to user input by allowing the user to manipulate the nodes which causes values to be set on the managed entity. This model can be made even more active by enhancing it with predictive capability. This enables the management system to manage itself, for example, to optimize its polling rate. The two major management protocols, SNMP  and CMIP , allow the management station to poll a managed entity to determine its state. In order to accomplish real-time and predictive network management in an efficient manner, the model should be updated with real-time state information when it becomes available, while other parts of the model work ahead in time. Those objects working ahead of real-time can predict future operation so that system management parameters such as polling times and thresholds can be dynamically adjusted and problems can be anticipated. The model will not deviate too far from reality because those processes which are found to deviate beyond a certain threshold will be rolled back, as explained in detail later. The process's messages must obey the rules for consistency in :
In order to determine the characteristics and performance of this predictive network management algorithm, we will review the research on performance and modeling of other lookahead algorithms and Time Warp in particular. In  a comparison of the conservative Chandy-Misra approach and the optimistic Time Warp is presented. This is done using a typical queuing theory approach which assumes exponential service times. There have been several other detailed comparisons between conservative and optimistic methods of simulation. These studies also make simplifying assumptions. In , it is shown that in a feed forward network, the time of execution of a message will occur earlier in Virtual Time than its corresponding message in the synchronous parallel algorithm described in . In , it is shown that Time Warp can out-perform the conservative technique known as Chandy-Misra by a factor of P, P being the number of processors, but that no such model in which Chandy-Misra out-performs Time Warp by a factor the number of processors used exists. Past work has examined the performance of Time Warp by comparing it to conservative mechanisms  or simulating the Time Warp mechanism itself . In this paper the focus is not only on analyzing and optimizing speed of execution but also using the algorithm to maintain network management prediction accuracy.
One goal of this research is to minimize polling overhead in the management of large systems . Instead of basing the polling rate on the characteristics of the data itself, the entity is emulated some time into the future in order to determine the characteristics of the data to be polled. Polling is still required with this predictive network management system in order to verify the accuracy of the emulation.