A CAC algorithm for per-VC queueing systems loaded by fractal traffic

ATM technology is expected to support a broad variety of services and applications, satisfying a wide range of user quality needs and network performance objectives. In this scenario a per-virtual connection (per-VC) queueing is needed to guarantee different quality of service (QoS) requirements and to provide adequate congestion avoidance techniques (such as early packet discard) to single traffic flows. We present a call admission control (CAC) scheme with per-VC management of network resources. Starting from the widely accepted long memory properties of broadband traffic, fractional Brownian motion (fBm) is used as a parsimonious traffic model. The proposed CAC scheme employs the knowledge of the exact asymptotic queue length distribution to efficiently allocate network resources (buffer and output link capacity) to the contending sources. We analyse the multiplexing gain considering an appropriate distribution of the bandwidth among different sources, highlighting its advantages over the rate envelope multiplexing scheme, suggested by the ITU-T