m0 -n6Z, Slides of Computer Networks

Download m0 -n6Z and more Slides Computer Networks in PDF only on Docsity!

COMPUTER NETWORKING Af STANFORD

suries m^0 -n6Z^

l-ol Fui.^ Tit^3

Dept. (^) oi Spe ;

• ,||^

'"i (^) YMO (^) Title (^).

"\

Computer (^) Networking (^) at (^) Stanford

Current Status and Future Plans for Ethernet at Stanford University

Ralph E. Gorin

Computer Science Department

Stanford University

January 1981

In the near future all major computing resources on the Stanford campus will be connected by the

Xerox Corporation's Ethernet communications network. This network will provide an unprecedented

level of system integration of immense value to the growing community of computer users. Many

new uses of computer systems, hitherto thought too inconvenient or too expensive, will become

practicable.

Our current^ efforts^ are^ aimed^ at connecting existing systems to Ethernet; these systems include

various configurations from several vendors, among which arc Xerox Corporation Alto personal

computers, a file server and a printing server; IBM 3033 and Series 1; Digital Equipment

Corporation PDP-11, VAX-11/780, and DECsystcms-10 and -20; and systems from Hewlett

Packard, Data General, and other vendors.

In addition to die progress being made connecting existing systems, we are designing Ethernet-based

stations (^) to support clusters of graphical or conventional (^) terminals. These terminal (^) systems will allow

users to communicate with any host computer on the Stanford University Network (SUN).

Portions of this^ document,^ particularly sections 3 and 4.3, are based on an unpublished proposal,

The SUN Workstation, [Baskett, Bechtolsheim].

m'

The word^ processing^ services planned by TNP (the Text Network Program in CIT) will be widely

accessible. Word processing will be used in the preparation of scholarly works, dissertations,

administrative documents, research reports, and proposals. Printing servers will be distributed to

convenient locations on campus; these servers will print files that are received from any network

host. Besides being able (^) to print data (^) and electronic (^) messages, these laser printers will (^) be able to

reproduce many different sizes and styles of typeface. Ordinary text, drafts of books, foriegn

languages, equations, and even simple drawings can be printed at the rate of eight or ten pages per

minute. As these printers will be (^) located in departmental (^) and adminstrative areas, (^) they will be

convenient to the people who use them. Classnotes, originals for transparencies, drafts for articles

and books, and all manner of printed material can be printed near each person's office..

Electronic mail will facilitate the day to day communication so necessary for the research and

management activities within the university. Casual use of electronic mail will replace some

memoranda; delays caused by the interdepartmental mail service will be eliminated. Professors and

adminstrators (^) will be able to (^) read their mail from home (^) terminals and while (^) traveling; messages can be entered into the system^ from (^) remote locations. Messages can be (^) annotated with (^) comments and

requests for clarification and returned to the originator or forwarded to other interested parties.

Database services will be important also. Portions of the library catalog will be available to each

person who uses the facilities attached to the network. Information necessary for financial and

academic planning will be available to the people' who need it Databases that are accessed and

updated through the network can make better information available to decision makers in a timely

manner and in an understandable format

The diversity of computer systems derives from the inherent superiority of particular systems for

particular tasks: not every system (^) is suitable for each possible application. (^) The (^) network allows (^) the

providers of computing services to select which services to make available on each different system.

The ability of a network terminal to access different systems de-emphasizes the need for all services

on every machine. Those systems tiiat (^) are suitable for word (^) processing will do word processing

instead of word processing, databases and statistics; machines suitable for databases won't have to

support matrix^ inversions,^ etc. The user of die network expends computer resources most efficiently

because each system where a particular task is performed has been tuned for that task.

As will be described below, Ediernet is particularly well suited to provide the basic network

services. The Telnet Protocol permits a terminal on any Ethernet host to engage in a terminal

session on any other participating host; die local host, the one to which the user's terminal is

directly connected, can be a personal computer, a timesharing server, or a special terminal

multiplexor, the Ether-Tip. Large amounts of data can be moved efficiently using the Ethernet File-

Transfer Protocol.^ Among^ the^ components^ needed^ to^ implement^ our^ pluralistic^ view^ of^ the^ future

of computing at Stanford^ arc Ethernet gateways, inter-network (protocol conversion) gateways,

Ether- tips, printing servers, and a low-cost, clustered display terminal system.

1.2 Selection of Ethernet

Early in^ 1979,^ as the Computer Science Department was planning its consolidation into one

building at Stanford, Margaret Jacks Hall, it was evident to many in the department that a critical

mass of^ computing systems^ was being assembled: there was no reasonable way to utilize these

4 4

different systems^ without^ having a network to interconnect them. Several of us then began to

survey (^) the state of networking technology.

In addition to the Ethernet, which we knew of from our contacts at Xerox PARC (Palo Alto

Research Center), we considered DECnct, the Chaosnet and LCSnet from MIT, and networks being

developed by MITRE and by BBN. Although appropriate for some environments, DECnet did not

appear appropriate for the network structure that we wanted for the department MlT's LCSnet is

not yet^ operational;^ the^ MITRE^ and BBN networking efforts are aimed at specialized environments

that are not characteristic of the situation at Stanford.

The Ethernet and Chaosnet both have the simplicity and flexibility that we sought. As Ethernet is

proprietary to Xerox, we did not expect to be allowed to use it. Chaosnet had no commercial

support: to use it we would have (^) to build most (^) componets. We (^) were considering, with (^) some

reluctance, the prospect of building our own Chaosnet when Xerox solved the Computer Science

Department's (^) problem by granting us equipment (^) and the right (^) to use (^) and augment (^) die current Ethernet.

The Computer Science Department's selection of Ethernet as die most appropriate technology for its

internal use has been ratified by the emergence of commercial support for the new Ethernet being

developed by Xerox,^ Digital Equipment Corporation, (^) and Intel: in the (^) near future (^) commercial (^) parts to (^) implement Ethernet will (^) be available, (^) as will be new systems that (^) are compatible with Ethernet

We believe that the Stanford Task Force on the Future of Computing should recommend the

adoption of Ethernet (^) as the standard for inter-computer (^) communications (^) on campus.

Adequacy of Ethernet for Future Applications

We recognize that in the long run both computing and communications requirements will expand.

At some point Ethernet^ will be inadequate, but we think that Ethernet has a reasonable life

expectancy. The 3 Mbit/second version of Etiiernet can handle about 375,000 characters per

second. Assuming that some fraction of these characters must be used for unavoidable network

overhead, we could still expect that some 300,000 characters can be transmitted each second. By

observation, systems such as the LOTS DECsystem-2060 can output between 5.000 and 10,

characters per second. Thus, one ether could support 30 to 60 machines of the size and speed of

the LOTS system. Each of these (^) systems supports 60 or more terminals comfortably, (^) so one ether might support between (^1800) and 3600 terminals.

We do expect that systems^ on the order of 10 times the speed of LOTS will be used at Stanford

during the (^) next five years. It will (^) be possible to completely (^) occupy the capacity of one ether. To

augment the capacity of the campus Ethernet, we will convert to the 10 Mbit/second version of

Etiiernet as soon as (^) practicable, and we will use (^) gateways to (^) obtain higher capacities by connecting several (^) ethers.

2. Ethernet in 1980

The Ethernet [Metcalfe] is a bit-serial, broadcast, multi-drop, packet switching network that allows

up to^255 stations (host^ computers or terminal concentrators) to be (^) connected via a single coaxial cable, (^) up to about (^) one mile long, (^) by simply tapping the wire. (^) The Etiiernet has (^) a bandwidth (^) of 3

By Protocol we mean a collection of software conventions used to describe a communications

process.

2.2. Advantages of Ethernet

Distributed Control: unlike other networks, such as ARPAnet or DECnet messages are not passed

through the^ hosts^ (or^ network processors) that are present between the source and

destination. Thus, the network continues to function even though some hosts may be

down with hardware, software or environmental problems. Moreover, there is no

centralized control for the network, thus eliminating one further source of unreliability.

(The gateway hosts are an exception to the passivity of intervening hosts. The purpose of

a gateway host is to pass messages from one network to another. Indeed, the loss of a

gateway may cause the network to become discontinuous, but a well-engineered network

can survive^ some such losses.)

Ease of Connection: the physical (^) connection to the ether is made through (^) a tap or sting, a simple

item of cable-TV technology. The sting can be installed while the ether is in use, without

causing any disruption.

Simplicity of the Medium: the ether itself is a standard co-axial cable. Connectors, terminators, T-

junctions, splices, etc. are all readily available. Co-axial cable is a more comfortable and

familiar (^) technology than the emerging fiber optic (^) devices. Although fiber optics will be

used in the future, few communications organizations are yet prepared to cope with a

phone call at 4:50 on Friday afternoon that announces that the fiber was just broken by a

jackhammer.

Simplicity of Protocols: the protocols, as described in section 3, are layered. Programs to implement

these layers can be well structured; they are easy to write (^) and can be organized in high- level languages that permit (^) the code to be (^) transported between different machines.

Elimination of Ad Hoc Interfaces: Seldom can a computer system be acquired without there soon

being a demand for it (^) to communicate with (^) other systems. (^) At Stanford, these (^) needs have caused a great profusion of special (^) purpose interfaces and software (^) to support them. Etiiernet promises (^) to eliminate (^) this difficulty: only (^) one new interface and one set of

software is needed for each new system. When a new computer can talk to Ethernet it

can talk to anyone.

Throughput: the present Ethernet at Stanford provides 3 Mbit/sec of nominal bandwidth, of which

some 85 to 92% is usable. This bandwidth is sixty times greater than any general

purpose links between computers at Stanford. The Etiiernet being developed by Xerox,

DEC and^ Intel^ will^ provide^ an additional factor of three greater throughput. The 10

Mbit/sec Ethernet is slightly faster than an RMOS disk. One 3 Mbit/sec ether will be

sufficient to^ support between^1800 and 3600 high-speed terminals; by using the 10

Mbit/sec version of Ethernet and by using gateways, we can support a very large number

of terminals and high bandwidth file transfers.

2.3. Overcoming Disadvantages

Limited Distance: Ethernet can be (^) extended to larger (^) areas' by (^) means of gateways (^) and inter-

network protocol conversion (to X.25, microwave, wide-band carriers, satellite links, etc.).

Limited Addressing Space: The address space of the present Ethernet 255 hosts per ether, will be

extended in the new Xerox, DEC, Intel version of Ethernet which provides 48 bits of

host ID.

Wiretapping and Privacy: Fundamentally, ethernet is a party-line situation. Mutually distrustful

parties can communicate via trap-door encryptions. Cooperating parties who don't trust

others can isolate themselves via encryption or by a private ether connected to the

outside via (^) a gateway.

Administrative Control: Any large organization already has a group that does nothing but deal with

die phone company. To expect mat less effort will suffice to administer a computer

communications network^ is^ not realistic. The evolving Ethernet technology should

include network management and evaluation tools, just as complex timesharing systems

have had performace evaluation tools built for them. Networking will not just happen:

as any other large undertaking, networking must be managed.

Software Effort: Every operating system that deals effectively with Ethernet must be trained in

Ethernet (^) protocols. This is inescapable. If (^) it is done (^) correctly, tiiis will be (^) die last time.

We intend diat future changes in transmission media will affect only the lowest level

driver software;^ new (^) network technologies (^) can be (^) adopted widiout (^) major disruption. (^) An

unintelligent connection to the Ethernet can be effected via the the Ether-Tip.

Limited Bandwidth: The Xerox ether is 3 Mbit/sec, sixty times the throughput of ARPAnet The

Xerox-DEC-Intel version of Ethernet is anodicr three times faster. When additional

diroughput is needed, gateways provide a means by which traffic engineering can effect

better use of bandwidth on each subnet. Eventually even tiiis technology will be

obsolete. Probably fiber optic ring networks will be die next media and architecture.

The great (^) hope for the future (^) is that media (^) changes should (^) be transparent to the software.

2.4. Current State of the SUN Ethernet

The curent status of the Stanford University Network is shown in figure 1. All the Xerox

equipment (16 Altos, the Dover printer, and the Ifs file server) and the three Vax systems are

fully functional Ethernet hosts. The Sail 1080 system is connected through the front-end PDP-11;

the primitive EFTP protocol has been implemented, but more complex operating system support for

Telnet and FTP has yet to be completed. Other machines are awaiting the fabrication of Ethernet

interface boards. These interfaces will connect the SCORE 2060 via the Massßus, die IBM 4331 via a

Scries/1 mincomputer, and other equipment such as the TI-990, via their own I/O busses or the

lEEE (^488) standard bus.

2.5. Plans for a Campus-Wide Ethernet

The current plans for the installation of Ethernet cable at Stanford calls for three phases:

Phase 1:^ Margaret^ Jacks Hall. (September 1979)

Phase 2: A "backbone" cable will run from the north-west corner of the main Quadrangle along

Via Crespi. From this cable, "spun;" with Ethernet repeaters will connect the Center for

Integrated Systems^ Annex, the Electronics Research Laboratory, the Applied Electronics

Laboratory, the Durand building, and the Terman building. (June 1980)

Phase 3: Cable spurs will be connected to the Medical Center Facility (Sumex), the Graduate

School of Business, the Center for Information Technology in Forsythe Hall, and the

Lots computer in^ the^ Ceras building. Ether-Tips will be added at various locations to

support the Text Network Program, CITs Terminals for Managers program, remote

access for Computer Science facilities, and for additional access to LOTS. (Commencing

first quarter^ of^ calendar^ 1981)

Sunet.fig2.sil

thernet, October, (^1980)

Proposed Extensions

/

2.6. Planning^ for Ethernet Growth

At the moment the Stanford Ethernet is an activity of the Computer Science Department and the

Computer Systems^ Laboratory^ of the Electrical Engineering Department These parties are involved

because they recognize their own urgent need for better inter-system communication; also they are

in a position where they (^) can actually implement (^) a network. In Computer Science (^) we recognize (^) the

network as both an object of research and as a necessary linkage between our various systems.

Looking ahead, the Computer Science Department has no desire to provide day-to-day operation

and management of a network service for the campus. We recommend that the campus-wide

computer communications network should be administered by the existing service center, CIT,

under the auspices of the Associate Provost for Information Systems. Naturally, the Computer

Science Department^ stands^ willing^ to help^ CIT in technical matters, but we must work towards the

goal of making the network a standard service offered by CIT.

The network services organization will have to address several problems. Among the problems that

must be resolved are

How will the capital costs of the network be financed?

How will the operating costs of the network be financed?

How (^) will the decisions that affect the (^) users of (^) the network be made?

We arc confident that these problems can be resolved in a manner that allows wide acceptance of

die computer (^) communications network.

EFTP exist in the C language under Unix and in assembler code for Pdp-11s. There is an ongoing

project between Stanford and CMU to implement all four Pup^ levels in C under Unix. This

software, along with MlT's C compiler for die MC6BOOO, will be used in our implementation of the

various Ethernet services. In addition, we are exchanging software with CMU, MIT, and Berkeley

on a regular basis. Stanford has developed software to print TgX, XGP, and Troff files on the

Dover. CMU and^ MIT^ are working on Pup^ packages and printing software.

Level 3 contains^ the^ user^ and^ server^ processes. They^ are: FTP User/Server for exchanging files,

Telnet User/Server to allow remote logins between hosts, and an EmPress style printing program

for the Dover. FTP and Telnet require RTP and BSP from level 2, while EmPress needs only

EFfP.

Level 4 is made up of specific terminal emulators and display packages. We can extend terminal-

dependent software to new terminals by means of these emulators.

Levels 4 and above

Level 3

Conventions for

data structuring and process (^) interaction

Level 2

Interprocess communication primitives

Level (^1)

Internet packet format Internet addressing Internet (^) routing

Level 0

Packet transport mechanisms

Application-defined protocols

Sunet.(ig3.sll

All hardware was designed with the SUDS design automation system. Most of the design was

initially expressed in (^) a graphical macro language (^) similar to Scald, but (^) more geared towards

efficient small-scale design. SUDS produces the wirelists for building wirewrap prototypes, and we

plan to use SUDS for laying out the PC boards for the pilot production run.

4.2 Configuration of the Ethernet Stations

The Ethernet Gateway (figure 4) is a dedicated processor that is equipped with two Ethernet

interfaces. The function of the gateway is to pass messages from one ether to the other.

Sophisticated routing algorithms (^) will handle complex (^) network topologies. Eventually, (^) we will

enhance the software to make protocol translation gateways that will interface to other networks such

as X.25, and to other physical transmission media.

A printing server is a dedicated station with device specific interfaces. One of the problems that a

network can solve is to make data appear near your office instead of at the computer center. The

advent of^ relatively low-cost highly flexible printers, such as the Canon LBP-10, makes it possible to

sprinkle printing stations throughout the campus. Moreover, these printing stations are not just for

traditional computer output: (^) they are quite appropriate for (^) use with the word processing systems

being used on (^) campus.

The Ether-Tip configuration (figure 5) can be used in locations where many conventional terminals

need access to Ethernet hosts. An Ethernet TIP also can be a central location for dial-up access to

all machines. Individual Ethernet hosts need not provide a variety of modems, line multiplexors,

and telephone^ equipment; dial-up access to all machines can be provided through one or more

Ether- Tips.

The Terminal Cluster configuration shown in figure 5 has eight high-resolution display channels.

The Tcnninal Cluster includes the Ethernet interface board, the microprocessor board, a graphics

controller and up to eight frame buffers. The (^) graphics controller and the frame (^) buffer for the high

resolution displays are described below.

The CRTeX/VLSI workstation is a station with a high-resolution display, keyboard, and tablet. For

VLSI design, we also want color graphics capabilities. When the hardware becomes available, we

plan to use the virtual MC6BOOO system with significant amounts of primary memory and a large

capacity disk for secondary storage.

4.3 High Resolution Graphics Terminals

We plan to use bit-map raster-scan graphics displays as output devices that provide a choice of

high-resolution monochrome, or medium-resolution greyscale or color.

A raster display is one of the most general output devices available today. It can represent

characters of arbitrary size and style, vectors and curves, solid and grey areas, and it can simulate

half-tone images [Thacker, Ncwman-Sproull]. A frame buffer display is also very economical: given

current dynamic memory costs of less than 0.05 cents per bit, a single-bit per pixel frame buffer is

about as expensive as die attached video monitor. As the price of dynamic RAMs continues to

decline, we expect to sec increasingly large frame buffers in the future, providing enhanced

16

resolution, grey-scale, and color.

Our design provides high-resolution displays with 1024X1024 pixels (or 1280X819). There is the

option of using lower resolution with multiple bits per pixel which can be mapped into colors or

grey-scale through a look-up table. The high-resolution display (^) is the configuration of choice for

design automation, text processing, and advanced program interfaces. It can display a high-quality

image of (^) a standard sheet of (^) paper in "portrait" mode.

Besides the resolution of a raster display, its update speed is a crucial factor. Our goal was to change

a complete high-resolution frame buffer without a noticeable delay. Specifically, the entire screen

should be scrolled within 64 milliseconds. The number of bits that must be accessed, shifted,

masked, and modified in that time requires significant processing power. We have developed a

novel frame^ buffer^ organization which reduces processing demands to a level where a single 16-bit

microcomputer can serve a number of frame buffers. In brief, a small amount of special hardware

implements a "RasterOp" rectangle manipulation function [Newman-Sproull] that makes it possible

to modify^ rasters in^ the^ frame^ buffer^ at full memory bandwiddi (32 Mbit/sec) without processor

intervention. The host processor only needs to set up the source and destination location, the height

and width of the rectangle, and the bit-operation desired. Excluding this overhead, a 16X

character can be put into the frame buffer in 16 microseconds, and a 1024X1024 RasterOp takes 64

milliseconds.

Sunet.figs.sil

References

[Baskett]

Forest Baskett,^ Pascal and Virtual Memory in a ZBOOO or MC6BOOO Based

Design Station, COMPCON 1980.

[Baskett Bechtolsheim]

Forest Baskett Andreas^ Bechtolsheim,^ Bill Nowicki, and John Seamons, The

SUN Workstation, unpublished proposal, March 1980.

[Boggs]

David R. Boggs,^ John F. Soch,^ Edward A. Taft, and Robert M. Metcalfe, Pup:

An Internetwork^ Architecture, lEEE Transactions on Communications, April

[Metcalfe]

Robert Metcalfe and David Boggs, Ethernet: Distributed Packet Switching for

Local Computer Networks, Communications of the ACM, volume 19 number 7

July 1976.

[Miller]

William F. Miller el al, Report of the Task Force on the Future of Computing

at Stanford University, to appear, 1981.

[Newman-Sproull]

William M. Newman and Robert F. Sproull, Principles of Interactive Computer

Graphics, 1979.

[Sproull]

Robert Sproull, Elaine Thomas, A Network Graphics Protocol, Computer

Graphics, volume 8 number 3,^ Fall (^) 1974.

[Thacker]

C. P. Thacker, E. M. McCreight, B. W. Lampson, R. F. Sproull, and D. R.

Boggs, Alto: A personal Computer, Computer Structures: Readings and

Examples (Siewiorek, Bell, and Newell, eds.) 1979.