Connection Admission Control Schemes for Self-Similar Traffic Yanping Wang Carey Williamson University of Saskatchewan
Connection Admission Control Features –important traffic management mechanism –improve network utilization (statistical multiplexing) –meet QOS requirements of all existing connections Difficulties –diverse traffic characteristics –various QOS requirements –bursty, self-similar traffic (i.e., LRD)
Self-Similarity:Prevalence of Bursts Over Many Time Scales
Properties of Self-Similar Traffic Autocorrelation Function
(continued) Variance-Time Plot
(continued) R/S Pox Diagram
Research Objectives Self-similarity –affects queuing behavior of aggregate traffic –has impact on network-engineering problems admission control, rate control Research objectives –CAC performance when presented with self-similar traffic –identify the impact of different parameters
CAC Algorithms PCR CAC –QOS guaranteed –network resource wasted SCR CAC –high network utilization –poor QOS performance AVG CAC –allocates extra bandwidth to handle the burstiness in the traffic GCAC –specified in P-NNI for efficient path selection –exploit multiplexing gains Norros CAC –based on FBM model –traffic characteristics captured by m, a, and H –exploit multiplexing gains
Experimental Methodology (1) Network Topology Fractional-ARIMA Based Model Hosking’s model + 3 transformations LRD and SRD features adjustable marginal distribution adjustable
Experimental Methodology (2) Simulation Configuration –ATM-TN simulator –factors (m, a, H, b, C and ) –metrics (CA, LU and CLR) –baseline configuration and the structured simulations Simulation Validation –warmup phase –accuracy of the results
Simulation Results (1): Baseline Configuration Call Acceptance Performance
Simulation Results (2): Baseline Configuration Link Utilization
Simulation Results (3): Baseline Configuration CLR Performance
Simulation Results (4): Baseline Configuration
Simulation Results (5): Parameter Effects Source Granularity
Simulation Results (6): Parameter Effects Source Variability
Conclusions (1): CAC Performance Impact of the Parameters –source granularity, source variability –long-range correlation structure –buffer size, target CLR –link capacity –mixing traffic sources Norros CAC and AVG CAC are promising –None of the CAC algorithms provides satisfying overall performance in all the scenarios
Conclusions (2): Impact of Self-Similarity Strong impact on network performance –especially when link capacity is small –statistical multiplexing gains should be exploited –achievable link utilization increases as link capacity increases –ineffectiveness of buffering
Future Work Multifractal property –multifractal vs. monofractal traffic Estimated traffic parameters –accurate vs. poor traffic parameters