Abstract: A plurality of multiprocessor systems is arranged in a high speed network to allow any processor in one system to communicate with any processor in another system. The network is configured as a multi-node dual bidirectional ring having a multiprocessor system at each node. Packets of information may be passed around the ring in either of two directions and are temporarily stored in buffer memory locations dedicated to a selected destination processor in a selected direction between each successive transfer between neighboring nodes. The buffer locations are managed so that they can request an adjacent node to stop transmitting packets if the buffer is becoming full from that direction and request resumption of transmission of packets as the buffer empties.

Abstract: It is becoming increasingly apparent that some aspects of intelligent behavior rcquirc enormous computational power and that some sort of massively parallel computing architecture is the most plausible way to deliver such power. Parallelism, rather than raw speed of the computing elements. seems to be the way that the brain gets such jobs done. But even if the need for massive parallelism is admitted, there is still the question of what kind of parallel architecture best fits the needs of various AI tasks. In this paper we will attempt to isolate a number of basic computational tasks that an intelligent system must perform. We will describe several families of massively parallel computing architectures, and we will see which of these computational tasks can be handled by each of these families. In particular, we will describe a new architecture, which we call the Boltzmann machine, whose abilities appear to include a number of tasks that are inefficient or impossible on the other architectures. FAMILIES OF PARALLEL ARCHITECTURES By “massively parallel” architectures, we mean machines with a very large number of processing elements (perhaps very simple ones) working on a single task. A massively parallel system may be complete and self-contained or it may be a special-purpose device, performing some particular task as part of a larger system that contains other modules of a different character. In this paper we will focus on the computation performed by a single parallel module, ignoring the issue of how to integrate a collection of modules into a complete system. * Scott Fahlman i* 3 supported by the Defense Advanced Research Projects Agency, Department of Defense, ARPA Order 3597, monitored by the Air Force Avionics Laboratory under contract F3361581-K-1539. The other two authors are supported by grants from the System Development Foundation. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Defense Advanced Research Projects Agency or the U.S. Government. One useful way of classifying these massively parallel architectures is by the type of signal that is passed among the clcmcnts. Fahlman (1982) proposes a division of thcsc systems into three classes: markerpassing, valuc-passing, and message-passing systems. Message-passing systems are the most powerful family, and by far the most complex. They pass around mcssagcs of arbitrary complexity, and perform complex operations on these messages. Such generality has its price: the individual computing clcmcnts are complex, the communication costs are high, and there may be severe contention and traffic congestion problems in the network. Message passing dots not seem plausible as a detailed model of processing in the brain. Such models are being actively studied elsewhere (Hillis, 1981; Hewitt, 1980) and we have nothing more to say about them here. Marker-passing systems, of which NETL (Fahlman, 1979) is an example, arc the simplest family and the most limited. In such systems, the communication among processing elements is in the form of single-bit markers. Each “node” element has the capacity to store a few distinct marker bits (typically 16) and to perform simple Boolean operations on the stored bits and on marker bits arriving from other elements. These nodes are connected by hardware “links” that pass markers from node to node, under orders from an external control computer. The links arc, in effect, dedicated private lines, so a lot of marker traffic can proceed in parallel. A node may be connected to any number of links, and it is the pattern of node-link connections that forms the system’s long-term memory. In NETL, the elements are wired up to form the nodes and links of a semantic network that represents some body of knowledge. Certain common but computation-intensive searches and deductions arc accomplished by passing markers from node to node through the links of this network. A key point about marker-passing systems is that there is never any contention due to message traffic. If many copies of the same marker arrive at a node at once, they are simply OR’ed together. Value-passing systems pass around continuous quantities or numbers and perform simple arithmetic operations on these values. From: AAAI-83 Proceedings. Copyright ©1983, AAAI (www.aaai.org). All rights reserved.

Abstract: A variable bandwidth branch exchange system is disclosed for interfacing a network ring to a plurality of peripheral loops, each of the peripheral loops being connected to one or more local stations and to a node on the network ring. Time bit slots on the network ring signal stream are assigned, on a dynamic basis, for communication between local stations. The network ring signal stream is diverted to the peripheral loop by the node when the bit slots assigned to the local station connected to that loop become accessible at the node, thereby placing the peripheral loop in the network ring signal stream. Bit slot bandwidth is variable in accordance with the requirements of a particular local station. Voice, data, and image communications are supported. The modular nature of the system permits the implementation of multiple rings for wide area networking.

Abstract: A two way digital communication arrangement utilizes a CATV system to provide bidirectional data transport service between any two points within the CATV system. The headend receives an upstream message and selectively rebroadcasts such message on the downstream portion of the spectrum. System intelligence is thus distributed throughout the system as server and subscriber nodes can be located anywhere in the CATV network. In order to obtain access to the CATV communication resources, user equipment at each node must attach a frame verifier (FV) code to each respective message. The headend examines the FV and permits rebroadcast of messages only if the FV code indicates that the user is authorized.

Abstract: A distributed problem solving network is a distributed network of semi-autonomous nodes that perform sophisticated problem solving and cooperatively interact with other nodes to solve a single problem. Because interaction among nodes is both limited and unreliable, each node directs its own activities in concert with other nodes, using potentially incomplete, inaccurate, and inconsistent information. Each node must balance its own perceptions of appropriate local problem solving activity with activities deemed important by other nodes. Organizational self-design is proposed as a multilevel approach to coordinating these networks. An organizational structure specifies the information and control relationships among the nodes in a general way. Each node is responsible for elaborating these relationships into precise activities to be performed by the node. This work focuses on the design and implementation of a coordination framework that is responsive to organizational structuring decisions. This framework is implemented as part of the distributed vehicle monitoring testbed: a flexible and fully-instrumented research tool for the empirical evaluation of distributed network designs and coordination policies. Each node uses a goal-directed Hearsay-II architecture and a local node planner to provide the sophisticated local control necessary for a node to evaluate its activity decisions based on internal, external, and organizational criteria. The capabilities of the framework are illustrated with testbed experiments using different organizational stuctures in four-node and five-node distributed networks.

Abstract: A data base management system in which data entities are records representing nodes (10, 12-15) in an antity-relationship directed graph. The body of each node represents one of the physical entities to be utilized while the edges (16-18, 21, 22) of each node represent relationships between that physical entity and other physical entities. Some of the edges are hyperedges (18) to permit the identification of simultaneous relationships with more than one other node. An application of this system to the assignment of telephone outside plant equipment to telephone subscribers is also described.

Abstract: The disclosed invention includes at least one programmable industrial measurement and control device operable in a stand-alone mode to sequentially and repetitively sample a plurality of analog input channels and store in preselected memory locations digital data representative of corresponding channels, to repetitively evaluate a predetermined plurality of industrial measurement and control functions on said data, and to actuate selected output channels in accordance with the results of each of the evaluation iterations to provide real-time control of an industrial process. Means including a pointer controlled ROM work list, a ROM program directory, a ROM variables control, a RAM report control, a RAM function control, and a ROM program code table specify the time sequence of, and the data and control structures for, each evaluation iteration. A host system is connected to the devices via a local area network for operation in a shared resources mode, and may be either a processor or an interactive terminal. The local area network includes two shielded twisted wire cables. A CSMA/CD communications protocol is implemented. Gateways connect the host to the network, and connect peripheral devices such as printers or graphics displays to the network. In the shared resources mode, each node on the network may routinely prepare and send reports to the host representative of sensed industrial conditions, may respond to host system generated commands, and may prepare and send exception reports to the host system whenever the industrial process exceeds prescribed bounds.

Abstract: A network of computing nodes is connected by communication channels into fully-interconnected units and groups of units. Each unit has one and only one direct connection to each other unit in a group, and any one computing node in any unit has a direct connection to at most one other unit. The nodes are suitably computing networks themselves so there can be any number of levels of recursion. In a method of routing and transferring information, each node is given an address having address portions respectively identifying its location at each level of recursion. The information is sent from that port of the sending node which has a port identification identical to the highest order address portion in the destination address which is different from the corresponding address portion of the sending computing node. Each computing node suitably has processing assemblies each having a digital processor, first and second memories, partitioning circuitry, and assembly control circuitry, communicating on a common intranodal shared bus. Each assembly is a port to an external bus. Each digital processor is able to communicate the information it generates from its first memory through the partitioning circuitry to a bus simultaneously with the partitioning circuitry also permitting access by other information to the second memory along the other bus.

Abstract: A communication system and method is provided, in which synchronous (i.e., clocked) serial digital data may be sent and received from any given node to any other given node along a multinode loop of any desired mode quantity, with each node being capable of and maintained ready to assume the role and function of master node to provide the time-base or master clock for the loop. One node will serve as master node and all other nodes as slave nodes until the master becomes inoperable in its master clock function or until it is removed from the loop, at which time another node will assume the role of master node, and this status will continue as above-indicated. Small loop size is accommodated by adding a suitable delay to retransmitted data at the master node. Each node has clock recovery and both recovered clock/data synchronization means and its on-board master clock/data synchronization means (which latter is close to the same frequency at each node, but independent in frequency and phase at each node) to enable each node to serve as either master or slave node by internal switching selection of communication control output of either recovered clock data or master clock data for use and retransmission at each node, dependent on its instant self-intended role as slave or master. Master clock data synchronization at the master node is effected by shifting recovered clock data by a selected phase as a function of phase difference between the instant master node master clock and recovered clock at such master node, the selected phase shift being an amount sufficient to enable effective sampling by the master clock, to thereby provide absolute phase synchronization of receive data with master clock for internal serial processing, utilization, and retransmission by the instant master node. Each instant slave node has its own on-board such master clock data synchronizing means which may be maintained on standby, for enabling each assumption of the master node role, as may be required.

Abstract: An arbitration technique for controlling access to a bit-serial bus by multiple nodes in a data processing network. Upon detection of no carrier on the bus (56), a node desiring access to the bus waits a predetermined number of quiet slots (60, 64), each slot being a predetermined interval. If that period elapses without another node's carrier being detected (64), the node desiring access is permitted to transmit (64, 68). For each node, two such delay-interval possibilities are provided, one high slot count (and, hence, low priority) and one low slot count (and, hence, high priority). The delay-interval selection for a node is switched from time to time on a round-robin basis so that all nodes get equal average priority. The high value of the delay interval is N+M+1 slots, where N is the node number and M is the maximum number of nodes allowed on the bus; the low value is N+1 slots. Initially, each node uses the former value. Upon unsuccessful contention for the bus, the delay-interval selection used next by the node depends on the number, LW, of the node which last won access to the bus. Upon detecting a carrier while waiting for access to the bus (i.e., losing arbitration to a higher-priority node), the node which is waiting for the bus compares its node number N to the number LW of the node which started transmitting (58). If LW was less than N, the node waiting for access uses a new waiting time of N+1 slots the next time the delay interval begins (62A); if LW was greater than N, the new delay interval value is N+M+1 slots (62B).

Abstract: A method of communicating data in a multinodal peer network communication system is disclosed in which the data is transmitted in accordance with an established priority scheme. The method uses two techniques to ensure that higher priority data is transmitted before lower priority data. In one technique, decreasing priority time-windows are established by each node of the system following the end of a transmission on a shared transmission medium. Subsequently, data is transmitted when the priority of the data matches or exceeds the priority of the time-window. In the other technique, data collisions on the shared medium are resolved by use of the priority time-window technique in combination with the use of a pseudo-random back-off timer.

Abstract: A communication method and packet switching system in which packets comprising logical addresses and voice/data information are communicated through the system by packet switching networks (116) which are interconnected by high speed digital trunks (118) with each of the latter being directly terminated on both ends by trunk controllers (131). During initial call setup of a particular call, central processors (115) associated with each network in the desired route store the necessary logical to physical address information in the controllers which perform all logical to physical address translations on subsequent packets of the call. Each network comprises duplicated arrays (170, 171) with each array having stages of switching nodes which are responsive to the physical address associated with a packet by a controller to communicate this packet to a designated subsequent node. Each trunk controller alternately transmits packets to available ones of the duplicated arrays as the packets are received from the attached trunk. If an array is unavailable for a time interval specified at initialization time by the associate processor, an affected trunk controller detects and transmits this fact to the associated processor via a maintenance message. Each trunk controller provides variable packet buffering and signaling protocols for each of the arrays in order to facilitate the transfer of packets from the arrays to the attached trunk.

Abstract: Several interconnection structures for a distributed multimicrocomputer message-passing system are compared on the basis of cost and performance. Among the structures analyzed are buses, double rings, D-dimensional toroids, trees, cube-connected cycles, and chordal rings. Network cost is defined in terms of the number of network nodes and the unit cost of communication links and their associated connections. Simple asymptotic performance bounds are derived based on the bottleneck analysis of a queueing network. In contrast to the usual assumption of uniform message routing, the technique permits the introduction of a reference locality notion to the message routing behavior of network nodes. Finally, the cost, performance, and performance/cost functions are examined as the number of network nodes becomes very large.

Abstract: Apparatus and a related method for regulating access to a communication bus to which multiple communication nodes are connected Control logic at each of the nodes determines which of them has priority to access the bus, by means of a parallel arbitration sequence in which all nodes contending for bus access participate Specifically, each contending node generates a relative priority node number and asserts it onto an arbitration bus All of the asserted node numbers are logically combined into a composite node number on the bus, and the winning node is determined in a bit-by-bit ripple comparison circuit at each node, the composite node number being compared with the locally generated relative priority node number Priority is determined in advance of data transmission, and synchronization and arbitration take place without any central or master control unit

Abstract: To preserve the consistency of on-line, long-lived, distributed data in the presence of concurrency and in the event of hardware failures, it is necessary to ensure atomicity and data resiliency in applications. The programming language Argus is designed to support such applications. This thesis investigates the mechanism needed to support the notion of data resiliency present in Argus. Data resiliency means that the probability is very high that the crash of a node or storage device in a distributed system does not cause the loss of vital data. Data resiliency requires the use of stable storage devices, memory devices that survive failure to a high probability. This thesis is not concerned with how to implement stable storage devices, but rather with how to organize the use of stable storage. The thesis presents a new organization of stable storage called the hybrid log that provides fast writing of information to stable storage and reasonably fast recovery of information from stable storage. In the context of this scheme, various algorithms are developed for writing objects to the log, recovering objects from the log, and housekeeping the log. {AD#A136484}

Abstract: A trunk controller and processor arrangement for monitoring the error rate occurring in packets received from a high speed trunk. Within a packet switching system, packets comprising logical addresses, and voice/data information are communicated through the system by packet switching networks which are interconnected by high speed digital trunks with each of the latter being directly terminated on both ends by trunk controllers. During initial call setup of a particular call, central processors associated with each network in the desired route store the necessary logical to physical address information in the controllers which perform all logical to physical address translation on subsequent packets of the call. Each network comprises stages of switching nodes which are responsive to the physical address associated with a packet by a controller to communicate that packet to a designated subsequent node. Each trunk controller has an error rate monitoring circuit for measuring the error rate occurring in packets during transmission over the attached trunk. The error rate circuit notifies the associated processor when error rate excursions increase or decrease in excess of a multitude of processor specified percentages of error rate.

Abstract: A multiple-ring communication system is disclosed wherein each node of the system is adapted to collect information on the status of the system. Receiver/transmitter equipments of each of the nodes are able to transmit ring test messages on the rings, the destinations of which are the node itself. Processing equipments control the receiver/transmitter equipments of the node and are able to check the receipt or absence of receipt of the ring test messages prior to possibly executing reconfiguration operations. Receiver/transmitter equipments of each of the nodes are also able to transmit neighboring node test messages on the rings, which upon receipt by the receiver/transmitter equipments of the neighboring nodes normally give rise to the transmission to the node of node test reply messages on rings different from those on which the neighboring node test messages were transmitted. The node processing equipments are also able to check the receipt or absence of receipt of the node test reply messages prior to possibly executing reconfiguration operations.

Abstract: A loop transmission system and a method of controlling the loop-back condition thereof. The loop transmission system comprises a plurality of node stations, a supervisory station, and two duplicate loop transmission lines which transmit signals in opposite directions. In the loop transmission system, the supervisory station first sends out loop-back commands via both of the duplicate loop transmission lines when faults are detected on both of the duplicate loop transmission lines at the same time. Each of the node stations establishes a loop-back path while retaining a connection path to a succeeding node station upon the receipt of a loop-back command. The loop-back path is released only in the node stations which receive signals normally from both of the duplicate loop transmission lines when the supervisory station sends release commands to the node stations after sending the loopback commands.

Abstract: Messages at nodes of a distributed data system are processed asynchronously. Each data atom shared by a pair of nodes is characterized by dipole halves maintained at each node including an access control characteristic, a quality control characteristic, and a pair of rank values. A rank value is a value which monotonically increases (with respect to time) with each significant event. A message from a paired node requesting a change to a dipole half at this node includes the set of rank values at the paired node for the shared data atom. This node compares the rank values which it maintains with those received from the paired node and, based upon the result, selectively rejects or processes the request.

Abstract: A video conversational data communication network (30) in which subscribers (34, 36) may conduct conversational video textual data communications with one or more keystations (70, 602, 98) in the network (30). Each keystation (70, 602, 98) is associated with a keystation terminal controller interface (68, 96, 600) which is in turn connected to a switching concentrator computer (46, 48, 110, 112, 114) and a message switching node (32, 42, 44) for routing calls throughout the network (30). The concentrators (46, 48, 110, 112, 114) enable calls to be directly routed to controllers (68, 96, 600) sharing the same node (32, 42, 44) without having to go through the node (32, 42, 44). The controller (68, 96, 600) locally stores (304, 306) video conversational textual data for its associated keystations (70, 82, 84, 98, 100, 602) and enables two different designated keystations to conduct two different video conversations with a common keystation in a split screen display (76). The split screen display (76) may also be used to display retrievable data from a data base (50, 52) for simultaneous display (76) along with a video conversation. The data is transmitted between connected controllers (96, 602) in packets which contain less than the total displayable data content of the video message input via the keyboard (72). The controller (96, 602) also enables preparation of responses prior to transmission to the other party and while receiving a transmission from that party. Prior to completion of a call, the controller (96, 602) provides an incoming calls queue video display (76) at the connected keystations (96, 602).

Abstract: Finite state automata have been applied with success to the modeling of Computer Network Protocols The interaction of finite state machines can be very complex especially if the protocol involves a large number of states To counteract the complexity of analysis and design, we propose an approach of decomposition Through this approach, the protocol graph can be partitioned into subgraphs each having a unique entry node and zero or more exit nodes The exit nodes of one subgraph can be connected only to the entry nodes of other subgraphs From the standpoint of protocol analysis, the correctness of the entire protocol graph can be inferred from the correctness of individual protocol subgraphs From the standpoint of protocol design, the individual protocol subgraphs can be designed to correspond to different phases of the protocol If the individual protocol subgraphs are designed correctly and the connections between subgraphs conform to the structure discussed above, then we show that the entire protocol graph will operate correctly

Abstract: This disclosure concerns a communications network information system, and ovides apparatus and a method for efficiently providing survivable communications capability to every surviving node in the communications network from every other surviving node. The types of information provided can include high priority message between users of the communications system, or perhaps of even greater importance, very frequently updated status information about the communications network nodes and their interconnecting links for use in optimally utilizing the surviving resources of a heavily damaged network. With this capability every surviving node in the network will be automatically and efficiently provided with the status of every node and interconnecting link.

Abstract: A data communications network has a transmission data channel wherein a plurality of network nodes are connected thereto according to a loop topology. Each node has a transceiver, a control switch for operating the node in either a pass through state, wherein received data from an upstream side of the channel is transmitted to the downstream side of the channel, or in a source transmit state wherein the node transmits its own message data to the downstream side of the channel while monitoring the upstream side. In the source transmit state, the node matches the received data with its transmitted data and truncates its transmission upon recognition of an error in the received message. In this manner, there is no requirement that the destination node acknowledge receipt of the data since the data received by the transmitting node is the most corrupted form of the data. The node further has a bus contention protocol wherein the node "backs off" from transmitting on the channel for a specific time-out duration. The time-out duration is varied depending upon the priority of the received message vis-a-vis the priority of the message being transmitted by the node. A node associated with a message having a higher priority "backs off" for a shorter time duration period. The time duration is determined by and at the node and can be programmed into a programmable, retriggerable, one-shot timing element.

Abstract: We recently described a method for measuring sinus node refractoriness in the rabbit heart. Atrial premature beats either may result in reset return responses or may become interpolated because of encroachment on sinus node refractoriness. In previous studies with rabbits we defined the effective refractory period of the sinus node (SNERP) as the longest premature interval that is interpolated. This study presents results on the extension of this technique to the measurement of sinus node refractoriness in man. Out of 30 patients (12 with and 18 without sinus node dysfunction), SNERP could be measured in 26 at one or more basic cycle lengths. At a basic pacing cycle length of 600 msec, SNERP ranged from 250 to 380 msec (mean 325 +/- 39) in patients without sinus node dysfunction and from 500 to 550 msec (mean 522 +/- 20) in patients with sinus node dysfunction. This clear differentiation of patients with and without sinus node dysfunction by SNERP is in contrast to various results obtained by assessing sinus node function from sinus node recovery time and sinoatrial conduction time. Thus this study suggests the possible use of the measurement of SNERP in the assessment of sinus node function in man and its possible value in identifying patients with sinus node dysfunction.

Abstract: The modeling and optimal flow control of a Jacksonian network in equilibrium is investigated. The model employed consists of a controller node cascaded with the Jacksonian network. Input packets arrive at the controller node with a Poissonian rate δ. For a blocking type strategy for accessing the network it is shown that the control which maximizes the average throughput of the network subject to a bounded average time delay constraint is a window flow control mechansim. The window size depends on the offered load δ, the maximum service rate of the controlling queueing system, c , and the Norton equivalent service rate of the network μ. The dependence of the average throughput and the average time delay on the control is also analyzed.

Abstract: In a system of the type in which processing units and terminal devices are connected to a common bus so that messages can be transmitted and received between these units and devices via the bus, the processing units and the terminal devices are connected to the common bus by node processors. A bus control method, the so-called "contention system", is adopted in which the first node processor starting a transmission of a message from the node processor to the common bus is given the right to transmit the message when a plurality of node processors have transmission messages to send. When transmissions are simultaneously started by two or more node processors, they are inhibited, and the messages to be transmitted are assigned priorities and are retransmitted by the corresponding node processors after the lapse of waiting times of different values which are set in advance in accordance with the priorities.