PDF of this tutorialBroadly speaking, the objectives of ATM traffic management are to deliver quality-of-service (QoS) guarantees for the multimedia applications and provide overall optimization of network resources. Meeting these objectives enables enhanced classes of service and offers the potential for service differentiation and increased revenues, while simplifying network operations and reducing network cost.
ATM traffic management and its various functions can be categorized into three distinct elements based on timing requirements.
First, are nodal-level controls that operate in real time. These are implemented in hardware and include queues supporting different loss and delay priorities, fairly weighted queue-servicing algorithms, and rate controls that provide policing and traffic shaping. Well-designed switch-buffer architectures and capacity are critical to effective network operation. Actual network experience and simulation has indicated that large, dynamically allocated output buffers provide the flexibility to offer the best price performance for supporting various traffic types with guaranteed QoS. Dynamically managing buffer space means that all shared buffer space is flexibly allocated to VCs on an as-needed basis. Additionally, per virtual connection (VC) queuing enables traffic shaping, and early and partial packet-level discard have been shown to improve network performance significantly.
Second, network-level controls operate in near real time. These are typically, but not exclusively, implemented in software including connection admission control (CAC) for new connections, network routing and rerouting systems, and flow-control-rate adaptation schemes. Network-level controls are the heart of any traffic-management system. Connection admission controls support sophisticated equivalent-bandwidth algorithms with a high degree of configuration flexibility, based on the cell rate for CBR VCs, average cell rate plus a configurable increment for VBR VCs, and minimum cell rate for ABR VCs. Dynamic class-of-service routing standards define support for fully distributed link-state routing protocols, auto-reconfiguration on failure and on congestion, and dynamic load spreading on trunk groups.
Flow control involves adjusting the cell rate of the source in response to congestion conditions and requires the implementation of closed loop congestion mechanisms. This does not apply to CBR traffic. For VBR and UBR traffic, flow control is left as a CPE function. With ABR, resource management (RM) cells are defined, which allow signaling of the explicit rate to be used by traffic sources. This is termed rate-based flow control. ABR is targeted at those applications that do not have fixed or predictable bandwidth requirements and require access to any spare bandwidth as quickly as possible while experiencing very low cell loss. This allows network operators to maximize the bandwidth utilization of their network and sell spare capacity to users at a substantial discount while still providing QoS guarantees. To enhance the effectiveness of network-resource utilization, the ABR standard provides for end-to-end, segment-by-segment, and hop-by-hop service adaptation.
Third, network engineering capabilities operating in nonreal time support data collection, configuration management, and planning tools (see Figure 7).
Figure 7. Network Engineering Capabilities
