WO1984004014A1

WO1984004014A1 - A distributed prioritized concentrator

Info

Publication number: WO1984004014A1
Application number: PCT/US1984/000347
Authority: WO
Inventors: Alan Huang
Original assignee: American Telephone & Telegraph
Priority date: 1983-03-28
Filing date: 1984-03-08
Publication date: 1984-10-11
Also published as: IT8420237A0; EP0138950B1; CA1207876A; KR920005106B1; EP0138950A4; ES531027A0; JPH0574977B2; DE3484413D1; JPS60500935A; EP0138950A1; IT1173759B; US4472801A; ES8501592A1; KR840008238A

Abstract

A distributed prioritized concentrator that manages the flow of information through high data rate communications paths. The concentrator accumulates incoming packets as necessary, sorts the accumulated packets in parallel based on the relative priority of the packets, and transmits as many of its highest priority packets as there are available communication channels. In its simplest form the prioritized concentrator comprises a parallel sorting network (30) having some of its inputs connected to accept incoming signal packets and the remaining ones of its inputs connected to outputs of delay elements (41). A number of the highest priority sorting network outputs are connected to outgoing communication channels (50), a number of lower priority sorting network outputs are connected to inputs of the delay elements (41) and the remaining number of lowest priority sorting network outputs are left unconnected. More sophisticated forms which add various capabilities are also disclosed.

Description

A DISTRIBUTED PRIORITIZED CONCENTRATOR

This invention relates to a prioritized concentrator having input and output lines. With the burgeoning of electronic technologies, the demand for a wider variety of communications services has been steadily rising. New technologies have been continually developing to meet these demands including lasers, fiber optics, microprocessors, and very large scale integrated circuits. While many alternative approaches for realizing the new communications services are available, digital data transmission seems to be the one best equipped for future progress and, accordingly, digital transmission is currently receiving the most attention. In terms of the services to be offered, it is expected that, in addition to standard telephone services, a multitude of low, medium and wide bandwidth services would be introduced. This would include electronic mail, facsimile, high-fidelity audio, computerized data base searches (e.g., want ads, encyclopedia, travel arrangements, etc.), remote shopping, electronic banking and home computer networks. Business users would also use electronic blackboards, teleconference facilities, word processors, and large computer communication networks. A number of approaches are employed in the prior art to manage the communication interconnections between central offices. One approach uses high capacity lines to interconnect central offices, but this is inefficient because most of the time some capacity is idle. The inefficiency is reduced in a variation on this approach which switches the information onto a small number of lines, taking advantage of the fact that not all of the lines are active simultaneously. This is the classic concentration function. It improves the utilization of the communication paths at the cost of possible blocking of signals. To reduce blocking of particularly vital signals, a third approach employs a computer to sequentially examine incoming lines, store the applied signal packets in a prioritized queue in the computer's memory and transmit the highest priority packets to its outgoing lines, as they become available. The difficulty with this approach is that a statistical multiplexer must scan the input, properly queue the data, and manage the outgoing lines in essentially a simultaneous manner. This becomes more and more difficult as the data rate increases or the number of incoming and outgoing lines increases. With present technology, the sequential nature of the computer approach cannot meet the demands imposed by the parallelism and consequent high "throughput" rates of the switching network we disclosed.

The problems are solved in accordance with the invention in a prioritized concentrator in which the concentrator comprises a sorting network having input and output ports for sorting signal packets applied to the input lines in accordance with a priority field accompanying the signal packets, the sorting network causing signals with successively higher priority to be connected to successively higher order output ports, and the sorting network having a number of its input ports connected to the concentrator input lines and another number of its output ports connected to the concentrator output lines, a delay network having input terminals and output terminals, and connecting circuitry for connecting the delay network input terminals to some of the remaining ones of the sorting network output ports and the delay network output terminals to remaining ones of the sorting network input ports. Summary of the Invention

To provide interconnection managing means for the high data rates of the switching network that effectively utilizes the available bandwidth of the communications paths among central offices, a distributed prioritized concentrator was invented. This concentrator accumulates incoming packets as necessary, sorts the accumulated packets in parallel based on the relative priority of the packets, and transmits as many of its highest priority packets to the next central office as there are communication channels to that central office.

In its simplest form the prioritized concentrator comprises a parallel sorting network having some of its inputs connected to output ports of a central office and the remaining ones of its inputs connected to outputs of delay elements. A number of sorting network outputs are connected to communication channels leading to another central office and another number of sorting network outputs are connected to inputs of the delay elements. More sophisticated forms which add various capabilities are, of course, possible and some of these capabilities are incorporated in this disclosure. Brief Description of the Drawing

FIG. 1 depicts a block diagram of a generalized distributed prioritized concentrator operating in acco rdance wi th the pr inc iples o f thi s invent i on ; and

FIG. 2 depicts a block diagram of a distributed prioritized concentrator which utilizes a sort and a merge network in place of sorting network 30 in the FIG. 1 system.

Detailed Description

The following description of the prioritized concentrator is that the signals described below flow in packets which contain a header field and a data field. The header field contains an activity bit (indicating whether the packet contains a valid signal - "0" -, or no signal - "1") and an address field including a destination address subfield and possibly additional subfields, with the most significant bit residing in first bit position of each subfield. For purposes of the prioritized concentrator disclosed herein a priority subfield is included in the header which specifies the packet's priority. Also for purposes of this disclosure, it is assumed that the priority subfield resides at, or has been moved to, the beginning of the header, immediately following the activity bit. A high binary number in the priority subfield has been selected to represent high priority and, conversely, a low binary number to represent low priority.

FIG. 1 presents a block diagram of a distributed prioritized concentrator embodying the principles of this invention. Sorting network 30 has a number of its inputs connected to a priority designation block 20 and the remaining ones of its inputs connected to outputs of a delay network 40. The inputs of priority designation block 20 are derived from the concentrator's input ports via a rate changing block 10. Priority designation block 20 assigns priorities to incoming packets when the concentrator interconnects central offices that do not assign packet priorities elsewhere. Rate changing block 10, as discussed below, is either a multiplexer or a demultiplexer, depending on the signal rates within and outside the concentrator. In some uses, neither block 10 nor block 20 would be required for the proper operation of the FIG. 1 concentrator.

Sorting network 30, whose output terminals are uniquely designated 0 through N-1 (starting with the left-most terminal in FIG. 1), sorts the packets applied to its input terminals based on the priority of the packets as specified in the priority subfield. That is, sorting network 30 routes the applied signal packets so that a packet at output terminal M has a priority equal to or lower than the priority of a packet at output terminal M-1, and a priority equal to or higher than the priority of a packet at output terminal M+1. Thus, the left-most output terminal of sorting network 30 in FIG. 1 receives the packet having the lowest priority. As shown in FIG. 1, a number of the highest designated output terminals of sorting network 30 are connected to a rate changing block 50 through which they are transmitted to the next star. Rate changing block 50, like block 10, is either a multiplexer or a demultiplexer depending on the relative data rates of the distributed prioritized concentrator and the outgoing lines. It, too, may not be necessary in some applications.

Most of the remaining packets at the output terminals of sorting network 30, having lower priorities and appearing at network 30 output terminals having lower designations, are applied to delay network 40 and are fed back to sorting network 30 input terminals. Each input in delay network 40 has a corresponding output in network 40, to which it is connected via a shift register 41. Each shift register 41 contains sufficient storage so that, in combination with the storage inherent in sorting network 30, exactly one packet is stored. This provides the appropriate feedback delay between the input to network 30, through delay network 40, and back to the input of network 30.

When network 30 has an equal number of input and output terminals, a number of sorting network 30 output terminals with the lowest designations are left unconnected and packets appearing at these terminals are lost. These packets are, in effect, overflow packets. Thus, at each iteration of sorting network 30 (at whatever iteration rate that is suitable for network 30) , the current highest priority packets are transmitted to the next star, most of the other packets are fed back to the input, and overflow packets, if any, are discarded.

The number of inputs and outputs of the concentrator, the incoming and outgoing packet rate and the iteration rate of sorting network 30 are all interrelated in a manner that may be appreciated from the discussion below. It yields a great deal of flexibility to this prioritized concentrator. When the iteration rate of sorting network 30 is

B and packets applied to the prioritized concentrator arrive on m₁ lines at rate n_1B, block 10 is a demultiplexer which converts m₁ lines at rate n₁B to m₁n₁ lines at rate B. Similarly, when outgoing packets leave the prioritized concentrator on m₂ lines at rate n₂B, block 50 is a multiplexer which converts m₂n₂ lines at rate B to m₂ lines at rate n₂B. Of course, m₂ is smaller than m₁ for, otherwise, there would be no need for a concentrator. The m₁n₁ outputs of block 10 are connected through block 20 to the input terminals of sorting network 30. Therefore, when network 30 has N inputs, N-m₁n₁ input terminals of network 30 are available to be connected to output terminals of delay network 40.

On the output side of sorting network 30, the m₂n₂ highest designation output terminals (carrying highest priority packets) are connected to rate changing block 50, the next N-m₁n₁ highest designation output terminals are connected to delay network 40, and the remaining N- (m₁n₁+m₂n₂) output terminals of network 30 are left unconnected.

From the above it may be observed that the size of sorting network 30, N, is the user's choice, as long as it is larger than m₁n₁+m₂n₂. - Larger values of N, with the associated larger memories contained in delay network 40, permit keeping a larger number of low priority packets before they are either transmitted or discarded via the unconnected output terminals of sorting network 30. It may also be observed that the prioritized concentrator can easily serve as a buffer between central offices that operate at different rates and, furthermore, that this ability is not dependent on any particular iteration rate within the prioritized concentrator.

Priority designation block 20 is shown in FIG. 1 following rate changing block 10, but it can precede it.

Its implementation depends, of course, on the particular scheme selected for assigning priorities. One scheme may simply comprise a "time stamp" that is applied to each incoming packet. To implement such a scheme, block 20 includes a digital clock and a register to connect each input line of block 20 to a corresponding output line of block 20. The size of each register equals the length of the priority subfield and, at the proper time, when the priority subfields of the packets traversing block 20 are in the registers, the clock time is inserted into the registers.

With respect to assigning priority to inactive packets which, of course, carry no information, the inverse of the activity bit may be used as part of the priority subfield. Active packet with activity bit "0" would automatically get a higher priority than inactive packets with activity bit "1". Alternatively, block 20 could detect empty packets (whatever their characteristics happen to be) and assign the lowest priority to those packets. Implementation of the various other blocks of

FIG. 1 may follow conventional design. For example, block 10 and 50 may be implemented with data selectors (e.g., TTL logic 74150 ICs) , data distributors (e.g., 74154 ICs) , and shift registers in a straight forward manner. Sorting network 30 may be implemented as taught by K. E. Batcher in "Sorting Networks and Their Applications", Spring Joint Computer Conference Proceedings, 1968, pages 307 et seq.

The structure of sorting network 30 is matrix like, with rows and columns and with a number of elements in each row, N, being equal to the number of input terminals of the network, and a number of rows that is monotonically increasing with N. Hence, the number of switching elements in network 30 is N times the number of rows.

The required number of switching elements can be reduced substantially when it is realized that the packets entering sorting network 30 from delay network 40 are already sorted. Capitalizing on this reduced entropy, sorting network 30 of FIG. 1 can be replaced with a much smaller sorting network that is followed by a simpler merge network. As shown in FIG. 2, sorting network 30 is replaced with a sorting network 70 that sorts only the m₁n₁ lines of block 20 (where m₁n₁<<N) and a merge network 60 that merges the output signals of sorting network 70 and of delay network 40. Network 60 is of width N but requires many fewer switching elements than a sorting network of the same width because its two input sets are already sorted. Implementation of merge network 60 can follow any one of standard approaches. One approach is described by Batcher in the aforementioned publication.

Claims

1. A prioritized concentrator having input and output lines;

CHARACTERIZED IN THAT the concentrator comprises: a sorting network (30) having input and output ports for sorting signal packets applied to the input lines in accordance with a priority field accompanying the signal packets, the sorting network causing signals with successively higher priority to be connected to successively higher order output ports, and the sorting network having a number of its input ports connected to the concentrator input lines and another number of its output ports connected to the concentrator output lines; a delay network (40) having input terminals and output terminals; and connecting circuitry for connecting the delay network input terminals to some of the remaining ones of the sorting network output ports and the delay network output terminals to remaining ones of the sorting network input ports.

2. A prioritized concentrator in accordance with claim 1 ,

CHARACTERIZED IN THAT the sorting network comprises a merge network having input and output ports, the sorting network output ports being connected to certain of the merge network input ports and the connecting circuitry connecting the delay network output terminals to other of the merge network input ports.

3. A prioritized concentrator in accordance with claim 1 ,

CHARACTERIZED IN THAT the delay network (40) comprises: a plurality of delay circuits (41) each of the delay circuits connecting an input terminal of the delay network, to an output terminal of the delay network.

4. A prioritized concentrator in accordance with claim 1 ,

CHARACTERIZED IN THAT the number of the sorting network output ports is smaller than the number of the sorting network input ports.

5. A prioritized concentrator in accordance with claim 1 ,

CHARACTERIZED IN THAT the concentrator further comprising: first multiplexing circuitry connected to each of the concentrator input lines for changing the rate of time division multiplexed signal packets on the input lines before applying them to the sorting network input ports; and second multiplexing circuitry connected to the another number of the storing network output ports for changing the rate of time division multiplexed signal packets on the another number of the output ports before applying them to the concentrator output lines.

6. A prioritized concentrator in accordance with claim 1 ,

CHARACTERIZED IN THAT the number of the sorting network input ports connected to the concentrator input lines is larger than the another number of the output ports connected to the concentrator output lines.

7. A prioritized concentrator in accordance with claim 1 , CHARACTERIZED IN THAT the concentrator further comprising: a priority controlling circuit interposed between the concentrator input lines and the sorting network input ports connected to the concentrator input lines.