Multiple Computer Networks and Intercomputer Communications

MULTIPLE COMPUTER NETWORKS AND INTERCOMPUTER COMMUNICATION

Lawrence G. Roberts Advanced Research Projects Agency Washington, D. C.

There are many reasons for establishing a executing a complex program using the data base. network which allows many computers to com- Use of a single data bank will save hardware re- municate with each other to interchange and exe- quired to store the information and will eliminate cute programs or data. The definition of a net- the need for maintaining multiple files. The term work within this paper will always be that of a "single data bank"does not necessarily mean the network between computers, not including the storing of only one copy of each basic data file. network of typewriter consoles surrounding each This type of use is particularly important to the computer. Attempts at computer networks have military for command and control, information been made in the past; however, the usual moti- retrieval, logistics and war gaming applications. vation has been either load sharing or interper- In these cases, one command would send a pro- sonal message handling. Three other more im- gram to be executed at another center where the portant reasons for computer networks exist, at data base existed. least with respect to scientific computer applications. Definitions of these reasons for a comput- Program Sharing: Data is sent to a program er network follow. located at a remote computer and the answer is returned. Software of particular efficiency or Load Sharing: Both the program and data capability exists on certain machines. For ex- are transmitted to a remote computer to equal- ample, if machine Y has a good LIST processor, ize the load between the two facilities. This type it may be more efficient for users whose local of operation requires identical computers or machine is X to use Y for LIST processing jobs. languages. When a given machine is loaded, con- Even if a LIST processor exists for X, the time sideration can be given to processing the program to execute the program on Y may be sufficiently on another machine. Many determinations must less than the time to execute on X that the total be made before an alternate machine is used (is time (and/or cost, including transmission) may there an alternate machine for which appropriate be less. The use of specialized programs at re- software exists, is that alternate machine in a mote facilities makes possible large gains in per- condition to handle the program, will more time formance. Perhaps even more important is the and dollars be spent on transmission than if the potential saving in reprogramming effort. user waits until the original machine is avail - able, etc. ). Such determinations are very diffi- Remote Service: Just a query need be sent cult and the gain only moderate, hence load shar- if both the program and the data exist at a remote ing is not a major consideration here. However, location. This will probably be the most common it is felt that some load equalization will occur mode of operation until communication costs come in any computer network. down. There will be a tendency for other cases to migrate toward this type of operation. For Message Service: In addition to computa- example, in a graphics application, the program tional network activities, a network can be used would be available or created on the remote com- to handle interpersonal message transmissions. puter and it would generate the data structure in This type of service can also be used for educa- its own computer. It would modify and update tional services and conference activities. How- the data structure from network commands trans- ever, it is not an important motivation for a net- mitting back display changes. This category in- work of scientific computers. cludes most of the advantages of program and data sharing but requires less information to be Data SharinG: The program is sent to a transmitted between the computers. remote computer where a large data base exists. This type of operation will be particularly useful The advantages which can be obtained when where data files are too large to be duplicated computers are interconnected in a network such economically. Frequently geographically dis- that remote running of programs is possible, in- bursed individuals need to access a common clude advantages due to specialized hardware and data base. Access to this data base may be re- software at particular nodes as well as increased quired simply to make an inquiry or may involve scientific communication. Specialized Hardware computer-to-computer link in such a network as It is felt that new machine configurations being the result of removing a user terminal can provide improvement factors of from 10 to from its cable on computer A, removing a user i00 in the problem area for which they were de- terminal from its cable on computer B, and signed. In some cases very large core and disk splicing the two computer cableends together. will substantially improve performance on exist- Such a network might operate as follows: (See ing machines. In other cases the improvements Figure 1) The user dials up his home computer, will be brought about by introduction of new sys- CA, from a console. He logs in normally by tems such as ILLIAC IV 1 and macromodular transmitting characters from his console to the machines. 2 A network is needed to make full use monitor. He sets up a user program and this of machines with specialized efficiency and with program, through the second channel, calls the a network the development of such computers remote computer, logs in, sets up the desired will be enhanced. user program on the remote computer, submits parameter data to it and receives the results. Specialized Systems Software Note that neither system was required to behave Handling jobs of widely varying sizes, par- in an unusual fashion. The monitors did what ticularly when initiated from many locations, they always do. The only requirement, as stated presents an extremely difficult scheduling prob- earlier, was that the user program be allowed to lem for any single machine. A large machine communicate with two terminals, its own user serving a number of smaller machines may pro- terminal and the remote computer. Most present- vide significant improvements in efficiency by day monitors provide for such a capability. alleviating the scheduling problem. Small time A computer-computer network link as des- sharing computers may be found to be efficiently cribed above was established in 1966 experimental- utilized when employed as communication equip- ly between MIT Lincoln Lab's TX-2 computer and ment for relaying user requests to some larger System Development Corporation's Q-32. 3 Both remote machine on which substantive work is nodes are general purpose, time-shared comput- done. What is envisioned is a system in which ers. This link allows programs on either com- the local machine serves some limited needs of puter to utilize programs such as compilers and the user while substantial requirements are graphics systems which exist only at the other satisfied by a remote computer particularly well node. The basic motivation was to test an initial adapted to handling the problem. network protocol, determine how well automatic dial up communications service worked, and Scientific Communication determine the extend of the time-sharing monitor Once it is practical to utilize programs at changes necessary. This has been done and the remote locations, programmers will consider link is now utilized by users to increase their investigating what exists elsewhere. The savings capability, thus providing more evaluation data. possible from non-duplication of effort are enor- mous. A network would foster the "community" Interface Message Processor. One way to use of computers. Cooperative programming make the implementation of a network between a would be stimulated, and in particular fields or set of time-shared computers more straightfor- disciplines it will be possible to achieve a ward and unified is to build a message switching "critical mass" of talent by allowing geographi- network with digital interfaces at each node. cally separated people to work effectively in This would imply that a small computer, an inter- interaction with a system. face message processor (IMP), would be located with each main computer to handle a communica- Approach tions interface. It would perform the functions of dial up, error checking, retransmission, routing Basic Operation. The minimum require- and verification. Thus the set of IMP's, plus the ment a system must meet to be eligible for telephone lines and data sets would constitute a membership in the network is a time-sharing message switching network (See Figure 2). monitor which allows user programs to communi- The major advantage of this plan is that a cate with at least two terminals. If this require- unified, straightforward design of the network ment is uniformly met, the network can be imple- can be made and implemented without undue con- mented without major change to the monitor at sideration of the main computer's buffer space, any installation, by the simple expedient of let- interpret speed and other machine requirements. ting each computer in the network look con- The interface to each computer would be a much ceptually upon all the others as though they were simpler digital connection with an additional flexi- its own remote user terminals. bility provided by programming the IMP. The Figuratively speaking, we may think of the network section of the I1ViP's program would be completely standard and provide guaranteed buf- location. Thus, given console type activity on fer space and uniform characteristics, thus the the network (messages of from l0 to 1000 char- entire planning job is substantially simplified. acters) the normal Z000 bits/second type com- The data sets and transmission lines utilized be- munication shou/d be sufficient at first. tween the IMP's would most likely be standard- ized upon, but as changes occurred in the com- Communication Needs munication tariffs or data rates, it would be more The common carriers currently provide 2 straightforward just to modify the program for or 4 wire, Z kc lines between two points either the IMP's rather than twenty different computers. dialed or leased, as well as higher band width As soon as the need became apparent, additional leased lines and lower band width teletype ser- small computers could be located at strategic vice. Considering the 2 kc offering, since it is connection points within the network to concen- the best dial up service, the use of Z wire ser- trate messages over cross-country lines. Finally, vice appears to be very inefficient for the type the modifications required to currently operating of traffic predicted for the network. In the systems would be substantially less utilizing Lincoln-SDC experimental link the average mes- these small computers since there would be no sage length appears to be 20 characters. Each requirement to find buffer spaces, hold messages message must be acknowledged so that the origi- for retransmission, verify reception of messages nator may retransmit or free the buffer. Thus and dial up telephone lines. the line must be reversed so often that the re- versal time will effectively half the transmission ARPA Net rate. Therefore, full duplex, four-wire service ARPA supports a number of computer re- is more economic and simpler to use. search groups throughout the country most of Current automatic dialing equipment re- which have their own time-shared computer quires about 20 seconds to obtain a connection facility. These researchers have agreed to and a similar time to disconnect. Thus, the accept a single network protocol so that they may response time is much too long assuming a call all participate in an experimental network. The is made only after a message arrives and that communication protocol is currently being de- the line is disconnected if no other messages veloped. It will conform to ASCII conventions arrive soon. It has proven necessary to hold a as a basic format and include provisions for line which is being used intermittently to obtain specifying the origin, destination, routing, block the one-tenth to one second response time re- size, and sum check of a message. Messages quired for interactive work. This is very waste- will be character strings or binary blocks but ful of the line and unless faster dial up times be- the communication protocol does not specify the come available, message switching and concen- internal form of such blocks. It is expected that tration will be very important to network partici- these conventions will be distributed in final pants. form during September 1967. Figure 3 shows a tentative layout of the net- BIB LIOGILAPHY work nodes and communication paths. However, i. Slotnick, D.L., "Achieving Large Computing since most of the communications will be dial-up, Capabilities Through An Array Compu- the paths are just hypothetical. It is hoped that ter, " presented at SJCC (April 1967). concentration and store and forward capability 2. Clark, W.A., "Macromodu/ar Computer will be available through the use of Interface Systems, " Proc. SJCC (April 1967), Message Processors. The development of the 3. Marill, T. and Roberts, L.G., "Toward a IMP's and the use of them at each node will allow Cooperative Network of Time-Shared store and forward operation as well as speeding Computers," Proc. FJCC (1966). the realization of a unified network. There are 35 computers shown in Figure 3 at 16 locations, there being several computers at most locations. A rough estimate would place the number of consoles attached to the 35 computers by the end of 1967 at 1500 and the number of displays at 150. Assuming four characters per second for typewriters and 20 characters per second for scopes, the total I/O rate to the computers is 9000 char/sec. Estimating that 10% of this I/O communication rate will be forwarded to another computer in the network leads us to an average transmission rate of 60 char/secper -.. ~. : . ' . . .;. ,-- r.. ~ " .. ".'..:.~,.:.,.., ....,. ~,, .: ...~. ,~,,",.~'~:.~','.'.,~~'.~i;.;~'-~:~ ,~:'~.: .,. ~ ".

• " '- " -. W ~v - :,.~:"~:~~i:::. " :.ii..;---

W~ Z3:

~ i,,,,1~,,,,,11 " Z Ww

.re,awl

' : .... t1,,,,,I'9t ' :. :? ": iI. ~" --. ~i~°~.,: , ~. .,~ ,":":i":,,,:. 0

' Z

v ~ Z

~ m z o~ ~- Z

u a O~ Z W ~

Z I z q~ im c/~J q~ Im

Z Inn 1,1,11 a~ CS~ I-- oa ~ 0q,.,p O~,n,.., O ~ I.n.i

um a. a. 0 =E z 0 0 I,a., I--- Z 1.1.1 Z aghhl hillv~ ,I~ CfJ RII IPnW~ Im q~ Z n.n.n I.-., :Z: CS~ 1.1.1 Z I--- IIIIIm I.n.I u.n.n C.~ ,ml _.i m,i I.n.I u--am ~ ~ mini Z Z n-- n-- ~ Z z ,i~ CS~J hhl hhl

hhl hhl ..I