Abstract: A data communications device interconnected for channelized communication overcomes the problems associated with message traffic starvation on outgoing traffic channels by forming arbitration, or speed groups of logical interfaces and allocating scheduling resources in proportion to the bandwidth attributed to each group relative to the total bandwidth of the device. Incoming messages have a transmission rate based on QOS, content type, or other constraints. A categorizer determines arbitration groups of the outgoing messages based on the transmission rate of the message. The scheduler allocates dequeue requests among the groups according to the proportion allocated to the arbitration group. A dequeue manager drains the outgoing messages from the arbitration groups at a rate in proportion to the total percentage of outgoing transmission bandwidth represented by the group by selecting a message from among the arbitration groups according to the computed proportion.

Abstract: Queues are scheduled for outputting data using a hierarchical tree. Each level of the hierarchical tree includes multiple entries for storing status information associated with the queues. For example, a given entry of the tree stores multi-parameter status information associated with a corresponding queue to be serviced. To determine which of the queues is next in line for servicing, a logic circuit compares the multi-parameter status information stored in two or more entries of the tree. The next scheduled queue or highest priority queue in the tree is then serviced by outputting at least some of its data.

Abstract: Typically, queues in a router device are used to track data information such as packets or data flows for later transmission to another downstream communication device. To reduce overhead associated with tracking an average queue length of a queue, a metric reflecting average queue length (or estimated queue length) is updated based on the occurrence of an event such as when the contents of the queue are modified. More specifically, if data is added or deleted from the queue, the metric reflecting an average fullness of the queue over time is updated at substantially the same time as updating the instantaneous queue length metric associated with the queue. This technique of synchronously updating both IQL and AQL simplifies the overall task of maintaining queue length information.

Abstract: A computerized device has dynamically modifiable hardware, such as an ASIC, that performs queue-scheduling operations. The hardware incorporates a generic sorting processor (GSP) that is dynamically configurable to implement various sorting algorithms to meet specific queue scheduling requirements for the computerized device. The computerized device extracts a first time stamp value and a second time stamp value associated with a first queue and a second queue, respectively. The computerized device receives instructions to configure a table of the GSP with scheduling entries. The computerized device compares the first time stamp value with the second time stamp value to form a comparison result. The computerized device then selects a decision instruction from the table, based upon the comparison result, and identifies a preferred queue of the first queue and the second queue, based upon the decision instruction.

Abstract: Conventional schedulers employ designs allocating specific processor and memory resources, such as memory for configuration data, state data, and scheduling engine processor resources for specific aspects of the scheduler, such as layers of the scheduling hierarchy, each of which consumes dedicated processor and memory resources. A generic, iterative scheduling engine, applicable to an arbitrary scheduling hierarchy structure having a variable number of hierarchy layers, receives a scheduling hierarchy structure having a predetermined number of layers, and allocates scheduling resources such as instructions and memory, according to scheduling logic, in response to design constraints and processing considerations.

Abstract: One or more queues store data information such as packets or data flows for later transmission to downstream communication devices. A real-time clock tracks current time and an advancement of a moving time reference, which is displaced with respect to the current time of the clock by an offset value. Thus, as current time advances, the moving time reference also advances in time. Upon servicing a queue, a time stamp associated with the serviced queue is also advanced in time. To monitor a rate of outputting data from the one or more queues, a processor device at least occasionally adjusts the offset value so that the moving time reference and values of the time stamps advance in relation to each other. Consequently, by tracking a relative time difference between current time of the real-time clock and a relative advancement of time stamps, a rate of outputting data information from the queue is monitored over time.

Abstract: A method and apparatus for performing logical multicasting within a network switch are disclosed so as to permit a downstream network device to perform cell forwarding in a manner which resembles spatial multicasting. A logical multicast cell has a connection identifier associated with the cell within the switch. The connection identifier is employed to initiate a series of lookup operations to generate unique VPI/VCI addresses which are transmitted out of an output port of the switch as copies or leaves of the multicast cell. Successive look-ups of outgoing VPI/VCI addresses for a logical multicast cell are performed using a chain bit to indicate when all leaves of the cell have been transmitted. Leaves of a logical multicast cell are transmitted to the downstream network device from one output cell buffer of a plurality of output cell buffers associated with the output port of the switch so as to minimize the likelihood of blocking and so as to minimize cell delay variation.

Abstract: A method and apparatus are disclosed for multicasting through a single physical port in a switch, i.e., logical multicasting. An incoming cell is logically multicast by providing a first copy of the cell to the desired output port and providing a second copy of the cell to a loop-back element which includes standard input and output ports dedicated to loop-back operation. The cell provided to the loop-back element is subsequently provided to the desired output port such that the first and second copies of the cell are serially transmitted from the single desired output port. Further, the cell provided to the loop-back element may be repeatedly copied, provided to the loop-back element, and provided to the desired output port. Thus, the incoming cell may be transmitted through a single output port as many times as is desired.

Abstract: Prioritized subclasses of cells within at least one quality of service (QoS) level in an asynchronous transfer mode (ATM) network element are described. Corresponding QoS queues are provided with one or more thresholds for intelligent discard of incoming cells within a QoS, based upon cell priority, at the onset of queue congestion. Cells having a first priority within a QoS are discarded as a first queue threshold is achieved; cells having a higher priority are enqueued. A cell loss priority (CLP) bit is employed to provide a further degree of cell discard prioritization. Different traffic streams within a single queue have different priorities, enabling intelligent discard of lower priority traffic prior to complete queue congestion.