A NEW PROTOCOL FOR A DISTRIBUTED LOOP ... - Springer Link

A NEW PROTOCOL FOR A DISTRIBUTED LOOP COMMUNICATION NETWORK

G. Papadopoulos, S. Leventis, S. Koubias Applied Electronics Lab. University of Patras, Greece.

ABSTRACT: This paper presents a new loop structure and its associated protocol for a mini-micro computer communication network. The proposed structure consists of two loops, both of which are used for control and data transmission one in the clockwise direction and the other in the counter-clockwise direction. This topology has all the advantages of ring structures such as simplicity, cost modularity, space modularity, concurrent service and high reliability. In addition, it provides better line utilization and traffic regulation than previously proposed loops because a) it has a distributed control mechanism and b) it allows two stations to communicate for long periods of time without blocking completely communication between other stations. Furthermore, it allows limited communication in case that .a break occurs in the line. The loop protocol is based on the SDLC frame format. This has made possible the use of the latest LSI technology that disasembles and reassembles the character of this frame. I.

INTRODUCTION

The objective of this work is to develop a network that is efficient, inexpensive and reliable and that will provide flexible service to a localized groop of users. Such a network must have the following properties: Cost modularity Space modularity Reliability Dynamic reconfiguration Universality 297

D. G. Lolniotisand N. S. Tzannes (eds.), Advances in Communications, 297-304. Copyright © 1980 by D. Reidel PubUshing Company.

G. PAPADOPOULOS ET AL.

298

High throughput Small installation cost There are many networks in use today, that have different topological structure. The most widely used structures are: a) Complete interconnection requiring N(N-1)/2 links for N nodes, b) Star configuration requiring N interconnections, c) Loop configuration, d) Bus connection. These structures have been examined by many researchers. The main conclusions with respect to the properties mentioned above are summurized in Table 1.

TABLE 1 Cost Modularity ---- ---- ---- ------- --Space Modularity -- ------ ----- - -- - - - --Installation Cost ------- -- ---- --- --- -Reliability - ------- --------------Distributed Control -- -- - - - - - ------- --- - -__ p.Y~~~~c__R_e~~n!ill~~ti°.!1__ __ l-~gic~I_~~m...P~~~Y.. _____ Throughput

INTERCONNECTION FULL

STAR

LOOP

BUS

No

No

Yes

Yes

Good

?

?

No Good

---- -- -- ---- -- --.:jt Yes ~ - No --- -------- ----;jt. Low 7i: lii[~ - - - - ---- ---------- ----- ------ ----Yes No Possible No ---- - ---- ----- 1----Yes No Possible No ----- ------ --- ----Ye s"" Yes --No-- -Yes - - - ----- ---->;Ie

trade-off

From this comparison it is clear that the full connected network, is too costly. The star network is less costly, has very good throughput but lacks in reliability, since, as soon as the central computer exhibits faulty operation, the whole system breaks down. The loop configuration seems to satisfy most of the desired properties and this is the reason we have addressed our attention to the design and construction of such a network. In addition, loop system are attractive for mini and microcomputer networks because of their simplicity and their high line utilization. Therefore, it was not surprising that loop systems have attracted the attention of many reasearchers who have designed a variety of loop networks (1), ( 2 ), (3), ( 4- ), (5), (6). The first loop system was suggested by Farmer and Newhall (1). In this system a round-robin control passing mechanism circulates around the loop and allows only one interface at a time to transmit one or more messages through the loop. During this transmission by one node, the other nodes have to wait. This results in a very inefficient loop channel utilization and long message delays.


299

In the Pierce loop (3) a time multiplexing mechanism is incorporated by dividing the communication space into one or more fixed size slots. In this scheme the messages are divided into packets, so that each packet will occupy one slot in the loop. This scheme improves the channel utilization, but wastes the communication space when a variable-length message is divided into fixed-size packets. Another disadvantage of this system is the extra hardware that must be designed for message disassembly and reassembly. It is apparent that neither the Farmer-Newhall or the Pierce loops make very efficient use of loop topology. Reames and Liu. (5) introduced a new message transmission mechanism called DLCN (Distributed Loop Computer Network) which combines the advantages of the Newhall and the Pierce loops. That is it allows multiple messages of variable length to circulate in the loop. This is done by inserting a variable length shift register before each node's transmitter. A condition that must hold is that the length of the locally generated messages cannot exceed the size of the delay buffer. Although the DLCN combines the advantages of the Farmer-Newhall and Pierce loops and also provides automatic traffic regulation (by means of the size of the delay buffer), it favors infrequent requests while delaying more frequent requests for network service. Other disadvantages of the DLCN are: a) b) c)

Interface complexity and, therefore, cost. Lower reliability due to the insertion of the variable shift register. The queuing time increases drastically with the number of nodes inserted in the loop.

In addition to the disadvantages discussed above, the previous loops have some additional common shortcomings. The most sign±ficar.t ones are: a)

b)

The stream of data is in one direction. Therefore, when a node communicates with the node just behind it in the loop, the transmission of data takes place through the rest of the nodes. This results, in more delay and lower reliability. A node can block out all other transmissions by sending a stream of messages to another node. This means that the above loops are sensitive to local demands which results in an over all decrease of the network performance.

With the motivation to the eliminate the above mentioned shortcomings a new loop structure was proposed by Jafari and Lewis (6). The main innovation in this loop is to distribute data and control into two different loops, a data loop and a controller loop. The data loop consists of bi-directional segments connect-

300


ing the nodes. Each node is interfaced to the loop by a zero-delay switch that may be turned "on" or "off". The operation of the control loop is based on a simple arbiter, which selects the minimum route. The structure of this loop permits concurrent data transmission between pairs of nodes connected by non-interfering segments. The zero delay relay nodes together with the capability of bidirectional message transfer between more than one pair of nodes improve considerably the throughput and the line utilization of this loop compared to the ones described earlier. However, ·the Jafari loop has a few disavantages of its own: a) b) c)

The control loop is under-utilized, since it remains idle for most of the time. Concurrent message transfer is allowed only between noninterfering nodes. The control mechanism is centralized.

In the loop structure to be proposed here, the two-loop concept is maintained. However, both loops are used for message transfer. One loop allows transmission only in the clockwise direction (called forward), while the other only in the counter-clockwisedirection (called backward). Thus, we can have messages of variable length transfered simultaneously between interfering node pairs. As far as the control operation is concerned, we have returned to the distributed control concept employed in the earlier loops. The distributed control operation together with the simultaneous bi-directional message transfer provide also automatic traffic regulation. This can be asserted on the basis of the following properties of the new loop: a) b)

II.

There is no central supervisor The previous individual accesses to the loop affect in an automatic way the passing of the control. DISCRIPTION OF THE NEW LOOP STRUCTURE AND PROTOCOL

As mentioned above the proposed network consists of two loops a forward and a backward. As shown in Fig. 1 the outside loop carries the forward traffic (clockwise), while the inside carries the backward traffic (counter-clockwise). Every node can become the controllerof the network as long as it has a message to transmit. Once a node gains control of the loop it establishes a forword communication link with the node to which it wants to transmit. The intervening nodes act as relays and their design is based on the SDLC chip (7), (8), which acts like a relay when it i? programmed in the one-bit delay mode. Two SDLC chips (one for each


CD node

- - -- - - IFg _

17 -

-...--

301

-

~controller

IF i-j forward transmission IF j-i backward 1/

FIG.1. The New Loop Structure

loop) under microprocessor control form the basic hardware for each node. The SDLC chip was found very convenient because it combines the relay, transmit and receive funtions all in one unit. The controller node makes first its identity known to all other nodes and starts transmitting the Information Frame (IF). The receiving node becomes itself a subcontroller, at the same time that it starts to receive the message. In this capacity it sends a Forward Subcontrol Frame (FSF) and a ~ackward Subcontrol Frame (BSF). The FSF gives the control to any other node beyond the subcontroller which wants to communicate with some other node in the forward direction but not past the main controller or with a node in the backward direction but not past the subcontroller. The node that made use of the FSF from the subcontroller acts now like a second controller. The node it communicates with acts like a secondary subcontroller. That is, the secondary subcontroller can transmit another FSF and BSF and so on. Thus, there can be more than one pair of secondary controller-subcontroller connections. From the above discussion it is clear that we can have simultaneous message transfer between many interfering node pairs. From

IF !A1!A2! F A1 A2 I FeS

:'

I FCS ! F ! ,

Opening, Closing Flag J 01111110. Destination Address Control: Source or End Address Information Field (variable leng.) Frame Check Sum

FIG .2. The Frame Format

IfIGIC~

I: s>~r:c~ : o

General Destination

1

2

7

11 I

/

Controller FIG.3. The Start Frame (SF)


302

an intuitive point of view this must increase considerably the line utilization over the Jafari loop. The communication protocol used in the proposed structure has the form of the SDLC protocol, the frame format of which is shown in Fig. 2, (7). The existence of chips developed for the implementation of this protocol made it particularly attractive for our application. The proposed method will best be described by an example, refering to Fig. 1. Let us assume that node 1 is the main controller and node 4 is the node that it wants to communicate with. That is node 4 is the first subcontroller. Also, let us assume that node 1 has been set in the transmit mode and all the others in the relay mode. Next, node 1 transmits a general frame that notifies all nodes about the identity of the main controller. This frame is called Start Frame (SF) and has the format shown in Fig. 3.

II

IF 1 0 S

O S

)(

I Jf

[ FCS ,

II

Destination Address Source Address

FIG.4. The Information Frame (IF)

F

"$FI F 10 FSF BSF

Ic I

~

I

i\

.

c

!FIG !Sb! F~SIF !Eopl Fr-

-I FIG I ElF QSIF IfOPI F I 012

7

Sb Iso u r c e I~bcontroller E I end addressl FIG.5. The Formats of FSF and BSF

After the transmission of the SF, node 1 starts to transmit the Information frame (IF) shown in Fig. 4. As soon as the subcontroller detects its address code in the destination field, it exits from the relay mode and enters the Receive Forward (RF) mode. At the same time it becomes the subcontroller and transmits two frames, one forward (FSF) and one backward (BSF), as mentioned in the beginning of this section. The formats of the FSF and BSF are shown in Fig. 5. It is seen that the subcontrol frames are followed by an EOP (End of Polling) character and a series of flags. The EOP character gives the opportunity to any of the nodes that it goes through to transmit an IF in the loop forward up to the main controller and backward up to the previous subcontroller. It is clear that he nodes that receive an FSF or a BSF become themselves in a sense traffic regulators. The format of the frame that is transmitted by a node k that receives an FSF and wants to transmit backward is shown in Fig. 6. It i~ seen that the EOP character that follows and FSF is converted a) to an opening flag for


303

a new FSF that notifies the nodes forward of k about the current status of the network and b) to an opening flag for the IF that is transmitted backward. ,

EOP (FSFlolll1,',l,ll FIFI F a) 10 1\ '. '. \ ,01 G !~~'I-! FyS ! F !EO~F b)-101'.'.1.1.10! D IS-=kl IF·J+- k S

FORWARD INPUT FS F 10,1,1,1,1,1,111

FIG.6. The response of a node k that receives FSF~ EO~ and wants to send IFj~k

FIG.7. The response of a node R. that receives FSF.. EO~and wants to send IFl...n.

)

}

(

-

, /

(

FORWARD OUTPUT

},I,l,111,l.oID=mIS:EI

IFe ....

m~ .

In Fig. 7 is depicted the response of a node £ that receives an FSF and wants to transmit a message forward. All the nodes forward of have received the FSF that went through node Z but do not find the EOP character that has been converted to an opening flag by node Thus, all the nodes forward of ~ up to node m that is supposed to receive the IF transmitted by f cannot transmit forward. The response of node m to the frame 1Ft ~ is shown in Fig. 8 •

e e.

PCF IFIGIGI

)

FSF BSF

I FIG I~fl F~S IF IEOpl-

-I FIG IE=~I FCS IF IEopl

FIG.8. The response of node m to

I~ ..... m

F~SIFlg!, . I •

• I

IFIG ICIF~S !F rI

SF

I

t

CONTROL TOKEN

FIG.9. The format of Pass Control Frame (PCF)

The above described process continues as long as there are available nodes in the loop. When the reception or transmission of a message at a node is completed, that node returns automatically to the one-bit delay (relay) mode, waiting for a new sequence of events to start. The new sequence is initiated upon the completion of the message transmitted by the main controller. As soon as this occurs, the controller will transmit forward and backward ABORT characters. These characters will return to the controller as long as all the other nodes have returned to the one-bit delay mode. Then, the


304

controller transmitts (PCr), followed by an for control take-over sage to transmit. In III.

forward the so-called PASS CONTROL FRAME EOP character. This gives the opportunity by the first forward node that has a mesFig. 9 the format of the PCF is depicted.

CONCLUSIONS

A new loop structure has been presented. This loop consists of two loops transmitting control and information packets in the clockwise and counter-clockwise directions respectively. It has been asserted that this new loop structure provides better line utilization than previously pr0posed loops. It does this while retaining the property of automatic traffic regulation in the sense that control is distributed and access to the loop is based on the current system load and previous individual accesses. The node hardware is based on the SDLC chip that handles the SDLC frames automatically, thus relieving the node processor of many tasks and reducing the required hardware. Preliminary investigation by the authors has shown that the proposed scheme is feasible. Hardware realization as well as analytic and simulation studies are currently underway. REFERENCES (1)

(2)

(3) (4) (5)

(6) (7) (8)

Farmer, W. W., Newhall, E. E., "An Experimental Distributed Switching System to Handle Bursty Computer Traffic", Proc. ACM Symposium. "Problems in the Optimization of Data Communications System", Pine Mtn. Georgia, Oct. 1969. Farber, D. J., Larson, K., "The Structure of a Distributed Computer System - The Communication System", Proc. Symp. on Computer Communications, Networks and Teletraffic, Polytechnic Institute of Brooklyn Press, 1972, pp. 21-27. Pierce, J. R., "Network for Block Switching of Data", BSTJ, Vol. 51, No.6, 1972, pp. 1133-1145. Fraser, A. G. "Spider - A Data Communication Experiment", Computing Science Technical Report No. 23, Bell Telephone Lab. Reames, C. C., Liu, M. T., "Design and Simulation of the Disstributed Loop Computer Network (DLCN)", in Proc. 3rd Annual Sympsium on Computer Architecture, Clearwater, Flori~a, January 1975, pp. 7-12. Jafari H., Spragins J., Lewis T., "A New Modular Loop Architecture for Distributed Computer Systems", Thrends and Applications: 1978 Distributed Processing. IBM Synchronous Data Control, Report No. GA27-3093-1 Intel Peripheral Design HandbookT~1978.