| Our research results are in the field of call or connection admission control (CAC) for computer communication networks. They are mainly related to connection-oriented packet switching networks, and address switch-level CAC, and what we call best-effort-friendly CAC and routing of connections with quality-of-service (QoS) requirements.; Best-effort-friendly CAC and routing of QoS connections refers to a methodology for protecting low-priority best-effort traffic in a network domain that provides both virtual-circuit routing with bandwidth reservation -- e.g., by Internet multiprotocol label switching (MPLS) -- for QoS traffic and datagram routing for best-effort traffic. We formulate a QoS-virtual-circuit admission control and routing policy that sustains a minimum level of best-effort performance. In response to a QoS connection request, the policy executes a two-stage optimization. The first stage seeks a minimum-net-effective-bandwidth reservation path that satisfies a best-effort protecting constraint; the second stage is a tie-breaking rule, selecting from tied paths one that least disturbs best-effort traffic. Our novel policy implementation efficiently executes both optimization stages simultaneously by a single run of Dijkstra's algorithm. Within a practical operating range, the consideration that our policy gives to best-effort service does not increase the blocking probability for QoS connection requests.; Switch-level CAC refers to the CAC that is performed for each link of a candidate path for supporting a QoS connection request. Associated with a link is its finite bandwidth capacity and, typically, an output buffer on the link's upstream switch. Packets are lost when the buffer overflows. We minimize the QoS connection requests' call blocking probability, as subject to a packet loss QoS constraint. For a single class of calls, we use our novel linear programming (LP) formulation, along with a convexity-like property and the complementary slackness condition from LP theory, to analytically prove optimality of a 'soft-threshold' CAC policy. This policy is a threshold policy in terms of the number of calls in progress using the link, with the qualification that the admission decision may be random at the threshold. We present the soft-threshold policy as an explicit algebraic expression of the problem parameters, and show its optimality for general conditions, including general call holding time distributions. |
