Eric G. Manning

On the progress of communication between two finite state machines

We consider the following problem concerning any two finite state machines M and N that exchange messages via two 1-directional channels. “Is there a positive integer K such that the communication between M and N over K -capacity channels is guaranteed to progress indefinitely?” The problem is shown to be undecidable in general. For a practical class of communicating machines, the problem is shown to be decidable, and the decidability algorithm is polynomial. We also discuss some sufficient conditions for the problem to have a positive answer; these sufficient conditions can be checked for the given M and N in polynomial time. We apply the results to some practical protocols to show that their communications will progress indefinitely.

Synchronization of distributed simulation using broadcast algorithms

Abstract Simulation, particularly of networks of queues, is an application with a high degree of inherent parallelism, and is of considerable practical interest. We continue the analysis of synchronization methods for distributed simulation, defined by the taxonomy in our previous paper. Specifically, we develop algorithms for time-driven simulation using a network of processors. For most of the synchronization methods considered, each node k of an n-node network simulation cannot proceed directly with its part of a simulation. Rather, it must compute some function B k ( ν 1 , ν 2 , …, ν n ), where ν i is some value which must be obtained from node i . The value of ν i at each node changes as the simulation progresses, and must be broadcast to every other node for the recomputation of the B -functions. In some cases, it is advantageous to compute the B -function in a distributed manner. Broadcast algorithms for such distributed computation are presented. Since the performance of a broadcast algorithm depends on the properties of the inter-process communication facility, we characterize some particular cases and give algorithms for each of them.

System Architecture for Distributed Data Management

Successful implementation of most distributed processing systems hinges on solutions to the problems of data mangement, some of which arise directly from the nature of distributed architecture, while others carry over from centralized systems, acquiring new importance in their broadened environment. Numerous solutions have been proposed for the most important of these problems.

A homogeneous network for data-sharing communications

Abstract The communications aspects of a distributed architecture for transaction processing are described. The architecture is aimed at transaction processing on physically distributed data bases, where most of the hits on a given component of the data base come from a single geographic region. The architecture is physically based on a homogeneous set of host minicomputers, a message-switched communications subnetwork (loop or packet-switched), and a set of network interface processors which connect the hosts to the communications subnetwork. It is logically based on two primitives; all data objects (including messages) are segments and all control objects (including files) are tasks. Each task runs in a private virtual space and all inter-task communication is done by passing message segments. Segment passing is done by a single message-switching task in each host, assisted by the interface processors and communications subnetwork where necessary. The message-switching task allso enforces protection rules without the need for special hardware. A two-host implementation of the logical architecture is operational. It is based on PDP-11 minicomputers and non-switched wire pair subnetwork. The companion paper describes modelling studies of the architecture, using simulation and queueing-theoretic techniques.

A computer architecture for large (distributed) data bases

It is argued that the data-base of a nation-wide organization will exhibit geographic locality of reference. That is, most of the transactions homing on a given component of the data base originate from a particular geographic region. At the same time there is a need to operate the collection of components as a single data base to provide for occasional transactions which cross regional boundaries, and for managerial queries and retrieval operations that span the entire data base. There are several examples of this associated with business and industry: credit and inventory records for exmple. Modest CPU power will suffice to perform most transaction processing on the data base. We are therefore led to consider a network of identical mini computers (or midis). Each host will execute an identical copy of the operating system in the network. The machines differ only in their complement of hardware (number of discs, primary storage size, etc.) and in the data they hold. Hence we assume that we are able to specify host hardware, host software and communications subnetwork as a single integrated system. The major goal of this work has been to investigate how far these freedoms can be exploited to yield simple elegant structures. The paper summarizes the design of the communication nucleus (4) of the network and focusses primarily on the design of the software to support transaction processing against the data base (3). Processor distribution is extended to include the ideas of a terminal host, a disc host, a central host and a communications device. The application of the proposed architecture to a typical commercial data processing problem is outlined.

On datagram service in public packet-switched networks

Abstract The need for datagram service in public packet-switched networks is examined. Also, the relationship of subnetwork design to datagram service is discussed. It is concluded that the use of layered design allows datagram service to be offered easily and efficiently, and also prevents the use of harmful routing and buffer allocation techniques.

A homogeneous computer network Analysis and simulation

Abstract A network of minicomputers to support transaction processing against a distributed data base is described. This paper gives a brief overview of the design based on a loop communications subnetwork and then describes analytic and simulation models that have been used to predict performance. Service demands in this network are assumed to have the following characteristics. Transaction: Users at on-line terminals enter messages that invoke short computations, a few accesses to the data base and a response message that is then sent back to the terminal. Locality of reference: Transactions entering at one node in the network can almost always be serviced locally, the fraction requiring remote service is expected to be in the 10% to 30% range. The network has been modelled analytically as a network queues. The predictions of this model have been compared to those of a large simulation of the network in an attempt to validate that simulation model. Agreement is good for a two host network over a broad range of network loading. The simulation program has been used to study a large number of problems where we varied: transaction characteristics, remote traffic fraction, host configuration and communications subnetwork speed. One such experiment is described here in detail.

Alternatives for interconnection of public packet switching data networks

CCITT has proposed Recommendation X.75 as a technique for interconnecting public data networks in a standard way. However, this is not the only or necessarily the best approach. This paper focuses on two major factors in determining the merits of different network interconnection approaches, and compares the X.75 approach to a few alternatives in each of these areas. One of the determining factors for interconnection is the willingness of each network to cooperate. X.75 requires a high degree of cooperation and uniformity of each networks operations. Another determining factor in interconnection is the degree of gateway participation in providing various levels of end-to-end service, such as virtual circuit and terminal handling services. The gateway may fully participate in and “terminate” the protocol for a given service, or may not participate in the protocol at all. The former leads to a stepwise implementation of an end-to-end service while the latter results in an endpoint approach. CCITT has chosen a largely terminating approach for virtual call service, but an endpoint approach for terminal handling. The degree of participation is examined for three alternatives: the X.75 gateway, the subscriber gateway and the datagram gateway.

An interprocess communication model for a distributed software testbed

This paper describes design and implementation of an IPC model for a distributed software testbed Shoshin. The IPC model was designed to provide a reliable message communication among distributed processes and to support implementations of user level protocols for various applications. A multiprocess structure, which separates a link level and a transport level, has been used to provide a flexible protocol management. The basic performance of the current implementation and the extensibility of the model are also discussed.

Protection and synchronisation in a message-switched system

Abstract In any computer system it is valuable to have methods for enforcing expected behaviour at runtime. (Capabilities and object-oriented architectures are mechanisms commonly used to address the problem.) In a message-switched system, all inter-process communication is done by the transmission of messages, and the problem of enforcing expected behaviour can be reduced to the problem of enforcing expected patterns of message transmission. In this paper we give a method, called the Task Graph Language, for expressing expected patterns (policy) of message transmission. Such an expression is quite independent of the actual programs which pass messages, and can be used at runtime to check the validity of observed behaviour. The mechanism we provide to perform the runtime checking of observed vs. expected behaviour is called Token Lists. This mechanism requires no esoteric hardware support and it can be used in both distributed and centralized systems. From another perspective, the user is provided with two independent, hence mutually redundant, ways to express his intent (the programs which pass messages and the Task Graph Language description), and the two expressions of intent are compared or matched at runtime as a checking mechanism. From still another perspective, the Task Graph Language and Token Lists permit a centralized representation of distributed control.