American Telephone & Telegraph Co., At&t Bell Laboratories

American Telephone and Telegraph, Co.

At&t Bell Laboratories

At&t Information Systems Inc.

Michael G Hluchyj

Yu S Yeh

Anthony Acampora

Mark John Karol

Kai Y Eng

Martin Braff

The present invention relates to a multichannel multihop lightwave communication network which allows packet concurrency in the network while avoiding the need for agile optical devices. The network comprises a lightwave communication medium (10) for supporting a plurality of independent user nodes; and a plurality of N network interface units (NIUs) (11). Each NIU is connected to the lightwave medium and one or more separate user nodes and comprises a trnasmitter section (28-35) and a receiver section (20-27) which are each fixedly assigned to transmit and receive, respectively, over a single channel or a plurality of separate channels to or from other NIUs of the network. In a connectivity pattern for one embodiment of the network, the NIUs are divided into k groups of p.sup.k NIUs such that each NIU of a first group of p.sup.k NIUs is capable of transmitting over the fixedly assigned single channel, or any or all of the p separate channels, to a preferred one or all of the associated p separate NIUs of a second group, etc. based on a generalization of a perfect shuffle pattern. Such technique permits all NIUs of the network to communicate with all other NIUs either via a single hop or via multiple hops where NIUs in subsequent groups of p.sup.k NIUs act as intermediate retransmission units.

Multichannel, multihop lightwave communication system

The present invention relates to an N-input, N-output 'Knockout' packet switch (11) which uses decentralized control and distributed routing. More particularly, within the switch, the N input signals comprising fixed-length packets propagate along separate broadcast buses (14) to each of N bus interface units (15) which include N packet filters (20), a concentrator (21) and a shared buffer (22). Each bus interface unit is associated with a separate one of the N outputs (12) of the switch, and the N packet filters therein are each associated with a separate one of the N broadcast buses for detecting if a packet on the associated bus is destined for the associated switch output. The concentrator is used to reduce the number of separate buffers needed to receive packets which may arrive simultaneously and are destined for the associated output. Contending packets at the output of the concentrator are stored in the shared buffer before placement on the output line on a first-in, first-out basis. The lost packet rate of the 'Knockout' switch can be made as small as desired and latency of packets in the switch is the smallest achievable by any comparable switch.

N-by-N 'knockout' switch for a high-performance packet switching system

The present invention relates to an N-input, N-output 'Knockout' packet switch (11) which uses decentralized control and distributed routing for routing high-speed, time-multiplexed, variable-length packets of information from the N inputs to the N outputs. More particularly, within the switch, the N input signals are separately synchronized such that the start of each arbitrarily arriving variable-length packet is synchronized to the start of a next mini time period of a sequence of mini time periods common to all inputs. The synchronized N input signals then propagate along separate broadcast buses (14) to each of N bus interface units (15) which include N packet filters (20), a concentrator (21) and a shared buffer (22) capable of processing variable-length packets. Each bus interface unit is associated with a separate one of the N outputs (12) of the switch, and the N packet filters therein are each associated with a separate one of the N broadcast buses for detecting if a packet on the associated bus is destined for the associated switch output. The concentrator is used to (a) reduce the number of separate buffers needed to receive packets which may arrive simultaneously and are destined for the associated output, and (b) only directs a maximum of L concurrently arriving packets to the shared buffer while discarding, or 'knocking' out, the remainder. Contending packets at the output of the concentrator are stored in the shared buffer before placement on the output line on a first-in, first-out basis. The discarded packet rate of the 'knockout' switch can be made as small as desired and latency of packets in the switch is the smallest achievable by any comparable switch.

A trunk circuit interfaces a multichannel data signal to a communication facility using a high priority queue (HPQ) memory, a low priority queue (LPQ) memory, and a data packet sorter including a separate LPQ counter for each data channel. Each LPQ channel counter keeps track of data packets from the associated channel which are stored in the LPQ memory. The packet sorter sorts data packets according to size. Small data packets from each channel which are smaller than a predetermined length are sorted for storage in the HPQ memory only when the associated LPQ channel counter is zero. Small data packets having a non-zero LPQ channel count and large data packets are stored in the LPQ memory. The trunk circuit first transmits data packets from the HPQ memory and then from the LPQ memory.

Growing community of inventors

Little Silver, NJ, United States of America

Michael G Hluchyj