The authors propose a congestion control framework for ATM networks that requires simple protocols and that provides guaranteed grade of service The authors suggest that the network provides two different services: express service and first-class service. Express service is appropriate for real-time applications, whereas first-class service is appropriate for nonreal-time applications. To provide such services, the proposed congestion control scheme prevents congestion inside the network by controlling the congestion at two levels: VC (virtual circuit) level and cell level. The VC-level control takes place during the call setup in the form of (logical) bandwidth reservation for individual VCs. The cell-level control is exercised across the network interface in the form of the choking/relieving of only first-class VC traffic in case of congestion. The authors analyze the performance of the proposed scheme, and show how to engineer various parameters to satisfy cell-level performance requirements (e.g., cell delay and loss).>

A congestion control framework for ATM (asynchronous transfer mode) networks is proposed. Specifically, it is suggested that the network provide two different services: express service and first-class service. Express service is appropriate for real-time applications (e.g., voice and video) whereas first-class service is appropriate for non-real-time applications (e.g., data). To provide such services, the proposed congestion control scheme prevents congestion inside the network by controlling the congestion at two levels: virtual circuit (VC) level and packet level. Express VC traffic is not subject to any flow control. At each intermediate node, transit packets are simply relayed with no traffic control. Various issues related to this scheme are discussed and its performance analyzed.>

The performance is analyzed of the congestion control scheme in signaling system no.7 (SS7). Unlike many other congestion control schemes used in data networks, a threshold control along with a choking mechanism is used in this scheme. Two analytic models are developed that are used to study the steady-state performance and the triggering behavior of the congestion control scheme. Both models are amenable to simple solution techniques. Numerical examples are provided that reveal the dynamics of the congestion control scheme with respect to various control parameters. From these results, a set of guidelines is elaborated for the determination of congestion control parameters that will result in good performance of the signaling network.>

A specific threshold load-balancing policy is described and analyzed for distributed computer systems in which each node executes job according to a priority schedule that depends on where the job originates. An approximate analysis is carried out to obtain the mean response time performance of the system in closed-form expression. In the analysis, it is assumed that the job transfers between nodes are Poisson processes. It is proved that the Poisson assumption on the job transfers is asymptotically exact (and hence the performance predictions of the model is asymptotically exact) as the number of nodes in the system increases. Simulations studies reveal that the Poisson job transfer assumption is good for small systems containing several nodes when one is interested in obtaining the response time performance of the system.>

The authors study the database query scheduling problem at STPs (signal transfer points) in a CCS (common channel signaling) network. For each arriving query, the STP can send the query to either one of the mated SCPs (service control points) to which it is connected. The authors consider a class of query scheduling policies ranging from policies that do not require any instantaneous workload information from SCPs to a policy that requires such information. Queuing models that describe the behavior of SCPs under each scheduling policy are developed and solved to yield the mean response-time performance of query processing. The authors carry out simulation experiments to obtain the second moment statistics of the response time performance. The results show that when SCPs provide enough concurrent processing capability, a simple round-robin policy yields a good performance as that of a more sophisticated policy requiring instantaneous workload information from SCPs.>

The authors develop distributed routing algorithms for large computer networks that provide virtual circuit connections to end users. One routing scheme in computer networks is to use a topology database maintained at each node that contains information about nodes and link cost. Using this information, each node finds shortest routes to other nodes. This scheme may be inefficient in large networks since the overhead associated with topology database update and shortest path computation may become large. One way of solving this problem is to partition the network into subnetworks or clusters. After partitioning, each node in a cluster only maintains a topology database of nodes and links in its own cluster and possibly marginal information about the entire network. For intracluster communication, shortest path routes are provided. For intercluster communication the source node collaborates with other nodes to find a path to the destination node. The authors propose two collaboration schemes that use simple protocols among network nodes. In one scheme, a broadcast search is used to find a path to the destination node in other clusters. In the other scheme each node maintains information about intercluster links on an up/down basis as well as complete topology database of its own cluster. Using this information, it finds a path to the destination node with the help of border nodes (nodes sharing intercluster links) in the intermediate clusters.>

Load balancing policies in distributed systems divide jobs into two classes; those processed at their of origination (local jobs) and those processed at some other site in the system after being transfered through a communication network (remote jobs). This paper considers a class of decentralized load balancing policies that use a threshold on the local job queue length at each host in making decisions for remote processing. They differ from each other according to how they assign priorities to each of these job classes, ranging from one providing favorable treatment to local jobs to one providing favorable treatment to remote jobs. Under each policy, the optimal load balancing problem is formulated as an optimization problem with respect to the threshold parameter. The optimal threshold is obtained numerically using matrix-geometric formulation and an iteration method. Last, we consider the effects that the job arrival process can have on performance. One expects that load balancing for systems operating in an environment of bursty job arrivals should be more beneficial than for an environment with random job arrivals. This fact is observed through numerical examples.