The problem of optimally choosing message rates for users of a store-and-forward network is analyzed. Multiple users sharing the links of the network each attempt to adjust their message rates to achieve an ideal network operating point or an "ideal tradeoff point between high throughput and low delay." Each user has a fixed path or virtual circuit. In this environment, a basic definition of "ideal delay-throughput tradeoff" is given and motivated. This definition concentrates on a fair allocation of network resources at network bottlenecks. This "ideal policy" is implemented via a decentralized algorithm that achieves the unique set of optimal throughputs. All sharers constrained by the same bottleneck are treated fairly by being assigned equal throughputs. A generalized definition of ideal tradeoff is then introduced to provide more flexibility in the choice of message rates. With this definition, the network may accommodate users with different types of message traffic. A transformation technique reduces the problem of optimizing this performance measure to the problem of optimizing the basic measure.

Bottleneck Flow Control

Packet switching offers attractive advantages over the more eonventional circuit-switched scheme, namely, flexibility in setting up user connections and more efficient use of resources after the connection is established. However, if user demands are allowed to exceed the system capacity, unpleasant congestion effects occur which rapidly neutralize the delay and efficiency advantages. Congestion can be eliminated by using an appropriate set of traffic monitoring and control procedures called flow control procedures. Flow control can be exercised at various levels in a packet network. The following levels are identified and discussed in this paper: hop level, entry-to-exit level, network access level, and transport level. For each level, the most representative techniques are surveyed and compared. Furthermore, the interaction between the different levels is discussed.

Flow Control: A Comparative Survey

iWarp is a system architecture for high speed signal, image and scientific computing. The heart of an iWarp system is the iWarp component: a single chip processor that requires only the addition of memory chips to form a complete system building block, called the iWarp cell. Each iWarp component contains both a powerful computation engine (20 MFLOPS) and a high throughput (320 MBytes/sec), low latency (100-150 ns) communication engine for interfacing with other iWarp cells. Because of its strong computation and communication capabilities, the iWarp component is a versatile building block for various high performance parallel systems. These systems range from special purpose systolic arrays to general purpose distributed memory computers. They are able to support both fine-grain parallel and coarse-grain distributed computation models simultaneously in the same system. An iWarp system can include a large number of cells; the initial iWarp demonstration system consists of an 8x8 torus of iWarp cells, delivering more than 1.2 GFLOPS. It can be expanded to include up to 1,024 cells. This paper describes the iWarp architecture and how it supports various communication models and system configurations.

/pdf/iwarp-an-integrated-solution-to-high-speed-parallel-3b7180n2p8.pdf

iWarp: an integrated solution to high-speed parallel computing

A computer communication network with window flow control is modeled by a closed multichain queueing network. The severe computational limitation of previous solution algorithms is overcome with a heuristic derived from the recently found mean value analysis. A large numerical example is given.

A Queueing Network Analysis of Computer Communication Networks with Window Flow Control

This paper first Surveys a number of potential deadlocks inherent in store-and-forward networks and outlines corresponding countermeasures. It then goes on to a more detailed treatment of the most important deadlock types. Finally, necessary and sufficient conditions are given and proven for the use of process ordering and generalized resource ordering techniques to avoid deadlocks in arbitrary systems of interacting processes.

E. Baubold

Papers

A Method of Deadlock-free Resource Allocation and Flow Control in Packet Networks.