Process Model/Queuing

Models of node architectures defined in the Node Editor may contain queue modules that represent packet buffering resources. An important aspect of modeling a queue involves specifying its behavior with regard to the method and timing of packet processing; this may involve such issues as the order in which to maintain the packets in the queue, the time at which to service the packets and how to select them, and the manner in which to use the multiple subqueues which a queue module may contain. In general, the process which manages a queue's resources and reacts to external events which may affect the queue's state is called the queue's discipline, in Riverbed a queue's discipline may be designated by the queue's "process model" attribute.

Queuing disciplines assigned to queue modules may be completely specified by a user. The mechanism used for this is the same as for general processes, namely the Proto-C language. Queuing process models make use of specialized kernel procedures which provide interfaces to the physical resources of the queue module. These are defined in the Q and Subq packages of the Simulation Kernel. By virtue of the fact that their behavior is specified in Proto-C, queues are fully programmable modules, allowing a wide range of queuing disciplines to be modeled.

Many queuing disciplines are provided as base models so that for common situations, the user can use the models already present in the /base directory, rather than define new ones in Proto-C. In addition, these models may serve as examples for the development of models of new queuing disciplines. In developing queuing discipline models, two broad classes of queues, called active and passive, are distinguished. Active queues are characterized by the fact that they autonomously dequeue and forward packets which they contain. They may elect to do so based on packet service times, or regularly scheduled servicing intervals, for example. In any case, they can be placed as an active component in a chain of processing elements, in the sense that no other element need request that the queue forward its contents. By contrast, passive queues do not have built-in service mechanisms. They will dequeue and forward packets only upon receiving some form of request from another module. This other module may be considered to be an external server for the queue. Requests to forward packets come in the form of access interrupts delivered by modules which are destinations of the queue's output streams. The destination modules cause these interrupts to occur by calling the Kernel Procedure op_strm_access().

The difference between passive and active queues is reflected in the names chosen for the queuing process models provided in the /base directory. Process models intended for use in active queues are named with a leading character 'a', and passive queues are named with a leading character 'p'. There are presently eight queuing process models provided. Each process model is described in individually in Model Architecture.

This chapter includes the following topics: