Google
 

Trailing-Edge - PDP-10 Archives - QT020_T20_4.1_6.1_SWSKIT_851021 - swskit-documentation/ddp.mem
There are 5 other files named ddp.mem in the archive. Click here to see a list.


+---------------+
! d i g i t a l !   I N T E R O F F I C E  M E M O R A N D U M
+---------------+

TO:  List
                                 DATE:  March 26, 1979

                                 FROM:  Peter Hurley

                                 DEPT:  LCG S. E. Dept.

                                 LOC:   MR1-2/E37  EXT:  6183

                                 FILE:  DDP.MEM

                                 PDM #: PMH-79-012-00-S


SUBJ:  DECSYSTEM-20 Distributed Processing Strategy

Distribution List

Dick Snyder
Larry Portner
Jack Mileski
George Plowman
Gordon Bell
Ulf Fagerquist
George Hoff
Per Hjerppe
John Jorgensen
Leslie Hruby
Bill Johnson
LCG Software Engineering Managers and Supervisors



     THE DECSYSTEM-20 DISTRIBUTED DATA PROCESSING STRATEGY



                           Abstract

The term "Distributed Data Processing" (DDP) is growing in popularity.
The  term is used for everything from dial up terminals and remote job
entry to a truly  distributed  database  where  interactive  inquiries
could  cause  accesses  to  data  stored  at  any node in the network.
However, most DDP systems have one major flaw,  they  are  limited  to
inter-system  links  of  56  K  baud  or  less.   The DECSYSTEM-20 DDP
strategy  not  only  covers   the   traditional   lower   speed   data
communication  facilities  but  it  also  encompasses high speed local
networks and multiple processor systems.  The  latter  facilities  are
aimed at greater ease of use, lower cost, and higher availability, all
of which are often sacrificed in the traditional DDP systems.

1.0  TRADITIONAL DISTRIBUTED DATA PROCESSING SYSTEMS. . . . . .  2
1.1  DECnet-20. . . . . . . . . . . . . . . . . . . . . . . . .  2
1.1.1  TOPS-20 Release 3A - . . . . . . . . . . . . . . . . . .  3
1.1.2  TOPS-10 Release 4 -. . . . . . . . . . . . . . . . . . .  3
1.1.3  TOPS-20 Release 5 -. . . . . . . . . . . . . . . . . . .  3
1.2  Other Traditional DDP Facilities On The DECSYSTEM-20 . . .  4

2.0  DEFICIENCIES OF THE TRADITIONAL DDP APPROACH . . . . . . .  5

3.0  LOCAL HIGH SPEED NETWORKS. . . . . . . . . . . . . . . . .  6
3.1  Ease Of Use. . . . . . . . . . . . . . . . . . . . . . . .  7
3.2  Lower Staff And Administrative Costs . . . . . . . . . . .  8
3.3  Growth Potential . . . . . . . . . . . . . . . . . . . . .  9
3.4  High Availability. . . . . . . . . . . . . . . . . . . . . 10
3.5  Performance. . . . . . . . . . . . . . . . . . . . . . . . 12
3.6  Hardware Requirements. . . . . . . . . . . . . . . . . . . 13
3.7  Restrictions . . . . . . . . . . . . . . . . . . . . . . . 14

4.0  20/VAX COMPATIBILITY . . . . . . . . . . . . . . . . . . . 14
4.1  20/VAX Growth Strategy . . . . . . . . . . . . . . . . . . 16
To:  List                                                       Page 2
Subj:  DECSYSTEM-20 Distributed Processing Strategy






1.0  TRADITIONAL DISTRIBUTED DATA PROCESSING SYSTEMS 


        The recently published Distributed Systems Handbook  describes
        distributed  processing systems as covering an extensive range
        of possibilities.  The range starts  with  distributed  access
        via dial-up terminals and remote job entry.  After distributed
        access comes distributed processing using  autonomous  systems
        without  any  communications  links.  In this environment each
        computer has its  own  data  base.   The  only  data  transfer
        between  these  computers  is  usually by magnetic tape.  This
        type of system is only classified  as  distributed  processing
        when  each  of  the computers is accomplishing some part of an
        overall objective.  The next step in distributed processing is
        to  link  the  computers  together  with communications links.
        However, because of the limited speed  of  the  communications
        lines  and  because of the complexity of designing distributed
        data processing applications most of the processing  and  data
        access  remains  local to each computer.  The ultimate form of
        distributed processing would be a network  of  computers  that
        share the data processing load and where data access to remote
        data bases is as easy as it is to a local data base.  However,
        there  are  very  few  of these systems commercially available
        today.

        IBM offers a large number of distributed processing  products.
        However,  these  are  limited  to functions such as remote job
        entry, remote job execution, file transfer, and  limited  file
        access.   A  recent  release  of  CICS/DOS/VS  introduces  the
        Inter-System Communications (ISC)  facility  which  gives  the
        user  the  ability  to access files and data bases on a remote
        CICS/VS  system.   Each  of  these  facilities  is  based   on
        traditional  communications  links  of  up  to 56 K baud.  The
        aggregate communications transfer rate is  also  very  limited
        (e.g.,  the  new  4331  processor  (.88 to .99, the speed of a
        370/138) can handle an aggregate rate of only 64 K baud).




1.1  DECnet-20


        DECnet-20 provides  the  basis  for  all  of  the  distributed
        processing  facilities  that are available on the DECSYSTEM-20
        family of computers.  The DECnet-20 program  started  in  1977
        with the birth of DECnet and it continues to closely track the
        corporate DECnet program.
To:  List                                                       Page 3
TRADITIONAL DISTRIBUTED DATA PROCESSING SYSTEMS 


1.1.1  TOPS-20 Release 3A - 

        DECnet-20 FCS

        DECnet-20 was first introduced  in  Release  3A.   This  first
        release  of DECnet-20 supported a single physical link between
        a DECSYSTEM-20 and any other Phase 2 DECnet system.   Although
        the  first  release of DECnet-20 was a complete implementation
        of NSP, only  task-to-task  communications  was  supported  by
        DEC-supplied   software.    Nonetheless,   this   has  allowed
        DECSYSTEM-20   customers   to   start   building   distributed
        processing networks.





1.1.2  TOPS-10 Release 4 - 

        Point-to-Point Connections

        Release 4 of TOPS-20 will be the second release of  DECnet-20.
        This  release will allow at least 4 physical connections to be
        made from a DECSYSTEM-20 to other DECnet Phase 2 (or Phase  3)
        systems.   Release 4 DECnet-20 will support any combination of
        point-to-point     connections.       Task-to-task      DECnet
        communications  can  be  established  between  any  two DECnet
        systems between which there is a direct physical link.

        File Transfer

        Release 4 will expand the DECnet-20 facilities to include file
        transfer  and  directory  control features such as listing the
        file names within a directory and deleting files.





1.1.3  TOPS-20 Release 5 - 

        Complex Topologies and Routing

        Release 5 of TOPS-20 proposes  to  support  even  more  DECnet
        facilities.   Specifically,  the DECnet-20 implementation will
        be upgraded to be a Phase 3 DECnet product.   DECnet  Phase  3
        will  support route-through and complex topologies.  This will
        allow  communications  to  be  established  between  any   two
        computers  in  the  network even though a direct physical link
        does not exist between them.

        Network Virtual Terminals

        Release 5 will also support Network Virtual  Terminals.   This
To:  List                                                       Page 4
TRADITIONAL DISTRIBUTED DATA PROCESSING SYSTEMS 


        facility  will  allow  users with access to a terminal that is
        physically connected to  a  DECSYSTEM-20  system  or  a  DN200
        remote  job  entry station in the network to logically connect
        that terminal to any other DECSYSTEM-20 in the network and use
        that  terminal  as  if  it  were  directly  connected  to that
        DECSYSTEM-20.  It is our intention that  the  Network  Virtual
        terminal  protocol  created  during  this  development  period
        become a DECnet standard.  Since  several  other  systems  are
        implementing  Network Virtual terminal support during the same
        time frame, we will be working with  those  groups  to  ensure
        that the implementations are compatible with each other.  This
        goal  is  extremely  important  in  our  effort   to   promote
        ease-of-use   across  all  of  DEC's  products  and  to  allow
        coexistence between DECSYSTEM-20 and VAX systems.

        Data Access Protocol (DAP) and RMS-20

        In order to increase compatibility between  files  created  by
        the  various  DECSYSTEM-20  languages as well as between other
        DECnet systems, Release  5  will  implement  the  Data  Access
        Protocol  (DAP).   The  basic  file opening and sequential I/O
        facilities of DAP will be supported by  the  TOPS-20  MONITOR.
        The  Data  Access  Facilities  of  DAP  will be implemented by
        RMS-20.  The GALAXY system will also be  enhanced  to  support
        the  submission  and execution of jobs to other DECnet systems
        in the network.

        Network Mail Facility

        Release 5 of TOPS-20  will  formally  support  a  DECnet  mail
        facility.    Although   this  facility  exists  today  on  the
        DECSYSTEM-20 ARPA system, it will be revised for Release 5  to
        conform  to  the  corporate  mail  protocols  currently  being
        specified.  This facility will allow DECnet users to send mail
        from one DECnet system to users on another DECnet system.  





1.2  Other Traditional DDP Facilities On The DECSYSTEM-20


        Remote Job Entry

        Release 4 of TOPS-20 will support a PDP-ll  based  Remote  Job
        Entry  station.   The  RJE station will be a DN200 with a line
        printer, a card reader, and a console terminal.  The DN200 RJE
        facility  will  be  enhanced  in  Release  5 to support Remote
        Terminal concentration as well as the line  printer  and  card
        reader.  The Release 5 DN200 RJE station will do local echoing
        and  editing  of  terminal  input.    This   will   make   the
        responsiveness  of  remote  terminals  close  to that of local
        terminals.  Terminals connected to the remote DN200 will  also
To:  List                                                       Page 5
TRADITIONAL DISTRIBUTED DATA PROCESSING SYSTEMS 


        be useable as Network Virtual Terminals.

        2780/3780 Emulation and Termination

        Release 3  was  the  first  DECSYSTEM-20  release  to  support
        2780/3780  emulation  and termination.  These products made it
        possible for the DECSYSTEM-20 to act as  a  Remote  Job  Entry
        station  on an IBM system, and for the DECSYSTEM-20 to receive
        jobs and print output on 2780/3780 Remote Job Entry  stations.
        This  facility  also allows disk data to be passed to and from
        IBM systems using the "Card Reader" and "Card Punch" devices.

        Hasp Multi-Leaving

        Release 4 of  TOPS-20  will  support  IBM  Hasp  Multi-leaving
        emulation  and  termination.   This facility is similar to the
        2780/3780 facility in that jobs  can  be  submitted  from  the
        DECSYSTEM-20  to  an  IBM  system or from a HASP Multi-leaving
        Remote Job  Entry  station  to  a  DECSYSTEM-20  system.   The
        Multi-leaving  protocol  allows  multiple  job  streams  to be
        active at once and allows the DECSYSTEM-20  operator  to  find
        out the status of the jobs that have been submitted to the IBM
        system.

        Gateways

        Starting with Release 5  we  plan  to  implement  Gateways  to
        several external networks.  Since the X.25 Gateway is the most
        important, Release 5 will implement  an  X.25  Gateway.   This
        will  allow  DECSYSTEM-20  systems  to  connect  to each other
        through an X.25 based network.  We will  also  be  looking  at
        supporting  the  X.25  terminal protocols so that terminals on
        the X.25 network can connect to a DECSYSTEM-20 .

        The knowledge and experience gained with X.25  will  lead  the
        way  for  other  gateways.   During  the Release 6 development
        period we will evaluate the requirements for  support  of  SNA
        and ACS networks.





2.0  DEFICIENCIES OF THE TRADITIONAL DDP APPROACH


        Over the next few releases the  DECSYSTEM-20  will  be  moving
        rapidly  into  the  traditional  Distributed  Data  Processing
        areas.  The  DECSYSTEM-20  will  support  interconnections  to
        DECnet  systems, IBM systems, and X.25 networks.  Users of the
        network software  will  be  able  to  transfer  files  between
        systems,  connect  their  terminals  to different hosts in the
        network, and access files on other systems in both  sequential
        and   random   access  modes.   The  structures  for  building
To:  List                                                       Page 6
DEFICIENCIES OF THE TRADITIONAL DDP APPROACH


        distributed processing applications will all be present.  

        Although the  plans  for  providing  a  base  for  distributed
        processing  on  the  DECSYSTEM-20  meet  all  of  the  current
        requirements for competing in this marketplace, there  is  one
        large  deficiency  with  this approach.  The data transmission
        bandwidths that are currently available are too low  to  allow
        large quantities of data to be transferred across the network.
        For example, table 1 indicates how long it would take to  copy
        a file containing the TOPS-20 monitor (330 pages, where 1 page
        = 512 36-bit words) from one system to another and how long it
        would  take  to  transfer  the data from one 2400 foot reel of
        tape (about 13000 pages) from one system to another.

             K Baud         Pages/Sec     File Transfer    Tape Transfer
        +---------------+---------------+---------------+---------------+
        !      2.4      !     .078      !    70.4 min   !   46.2 hours  !
        +---------------+---------------+---------------+---------------+
        !      4.8      !     .156      !    35.2 min   !   23.1 hours  !
        +---------------+---------------+---------------+---------------+
        !      9.6      !     .312      !    17.6 min   !   11.6 hours  !
        +---------------+---------------+---------------+---------------+
        !     19.2      !     .625      !     8.8 min   !    5.8 hours  !
        +---------------+---------------+---------------+---------------+
        !     56.0      !     1.82      !     3.0 min   !    2.0 hours  !
        +---------------+---------------+---------------+---------------+
                Time to transfer a 330 page file or a 2400 ft tape
                          (assumes 60% line efficiency)

                                TABLE 1

        It should be quite evident from this table  that  transferring
        large  amounts  of  data  over  the synchronous communications
        lines  takes  exorbitant  amounts  of  time.   Therefore,  the
        current   distributed   processing   systems   have  all  been
        restricted to doing most of their work locally with occasional
        data transfers between systems.




3.0  LOCAL HIGH SPEED NETWORKS


        The  traditional  distributed  data  processing  approach   is
        definitely  required  for those instances where the systems in
        the network are geographically far apart and where the network
        is  composed  of  heterogeneous systems.  However, there is an
        alternative approach that has not been  fully  exploited  yet.
        In  the  cases  where  the  network  of  computers  is  highly
        localized (e.g.  the same building or cluster  of  buildings),
        then  a  high speed local network provides many benefits.  The
        DECSYSTEM-20  distributed  processing  strategy  proposes   to
        provide   both   the  traditional  form  of  distributed  data
To:  List                                                       Page 7
LOCAL HIGH SPEED NETWORKS


        processing and a local high  speed  network  capability.   The
        local  network  will  consist  of  a  homogeneous  network  of
        DECSYSTEM-20 computers.  The network will not  be  limited  to
        just  one  type of DECSYSTEM-20, but any mix of computers from
        the entire DECSYSTEM-20 family (provided that the  appropriate
        hardware  interconnects  are  built).   The  high  speed local
        network is being designed to provide our customers with growth
        potential,  ease  of use, lower cost, and higher availability.
        It should be remembered that the local network being presented
        here   is   applicable   only  to  a  homogeneous  network  of
        DECSYSTEM-20s and that the traditional distributed  processing
        facilities  such  as  DECnet, IBM 2780/3780 and HASP, and X.25
        probably  represent  a  larger  and  more   important   market
        especially  in light of the recent growth of networking in the
        computer industry.





3.1  Ease Of Use


        The DECSYSTEM-20 local high speed network is being designed to
        provide  the  greatest  possible  ease  of use.  The following
        features will be a basic part  of  the  design  of  the  local
        network.

        File System Transparency

           The fact that the computer facility is made up of a network
           of  separate  systems will be transparent to the user.  The
           user can log into any system in the network and access  all
           file structures in the network.  The file structures within
           the network will all  be  named  with  network-wide  unique
           names   (however,   this   is  not  a  fixed  requirement).
           Therefore,  node  names  need  not  be  included   in   the
           specification of a file.  This also allows structures to be
           transparently moved from system to system thereby providing
           higher  availability  in  case of hardware failures.  Users
           will be able to run all of the DEC supported languages  and
           applications  without modification even though the files or
           data  bases  being  referenced  are  on  disks   that   are
           physically connected to other systems in the network.  This
           includes doing simultaneous access and  update  to  network
           wide  data bases.  Users will be able to "CONNECT (to)" and
           "ACCESS" directories on any file structure in  the  network
           (subject to the normal protection mechanisms) regardless of
           which system the directories reside on.

        Network Wide ENQ/DEQ Facility

           The  ENQ/DEQ  facility  that  is  offered  on  the   single
           processor  DECSYSTEM-20 will be expanded to support network
To:  List                                                       Page 8
LOCAL HIGH SPEED NETWORKS


           wide use.  This means  that  applications  that  have  been
           written  to  allow  simultaneous  file access using ENQ/DEQ
           will run without change in the local network environment.

        Network Wide IPCF and INFO

           The local network will support inter-system messages  using
           the  Inter-Process  Communications  Facility (IPCF).  There
           will be a network wide INFO process for signing out process
           names  and  for  determining the PIDs of other processes in
           the network.  

        Increased Programmer Productivity

           Because  the  programming  habits   of   the   applications
           programmer  does  not  have  to change to write distributed
           processing  applications  for  the   local   network,   the
           productivity  will be increased.  Likewise, the ease-of-use
           qualities of the network will lower the number of  bugs  in
           these local network applications.





3.2  Lower Staff And Administrative Costs


        One of the primary goals of the local network  project  is  to
        reduce the cost of running the network.  In general, the staff
        and administrative costs associated  with  operating  a  local
        network will be significantly lower than the cost of operating
        each system separately.  To achieve this goal,  the  following
        facilities will exist.  

        Central Operator Work Station

           From  a  single  terminal  anywhere  in  the  network,  the
           operator will be able to perform all of the normal operator
           functions.  These include  controling  the  line  printers,
           card  readers, disk and tape mount requests, user requests,
           the batch streams, and the external network links.  

        Central Accounting 

           The accounting and billing facilities will be  centralized.
           The  accounting  data from each system can be collected and
           processed  together.   This  provides  as  much  uniformity
           across the network as is desired.

        Central File Backup and Archiving

           The disk backup and  archiving  procedures  for  the  whole
           network  can  be  performed  using  a central tape library.
To:  List                                                       Page 9
LOCAL HIGH SPEED NETWORKS


           This  simplifies  the  management  of   the   file   system
           immensely.   Since  the  file  system  backup and archiving
           activity can be performed on a central set of tape  drives,
           the  number  of tape drives required in the network will be
           substantially lower.  Likewise, the number of reels of tape
           required  to  maintain  the  network's file system archives
           will be reduced.

        Central Spooling and Shared Devices

           The  line  printer  and  card  reader  spooling   will   be
           centralized.   This means that fewer line printers and card
           readers will be  required,  lowering  the  equipment  costs
           significantly for the smaller systems in the network.  

        Central Access Control

           The  access  control  algorithms  set  up  by  the  network
           administrator  can  be  common  throughout the entire local
           network.  This reduces the  amount  of  programming  effort
           required  to  support  the  individual  systems.  This also
           provides  a  uniform  level  of  security  throughout   the
           network.





3.3  Growth Potential


        The local network facility will allow  DECSYSTEM-20  customers
        to  incrementally  expand  their computing facilities until it
        reaches about 6 to 10 times the capacity of the largest member
        of  the  DECSYSTEM-20  family.   Assuming  that  the  required
        hardware links are built,  customers  will  be  able  to  link
        together  any type of computer from the DECSYSTEM-20 family in
        any  combination.   Depending  on  the  requirements  of   the
        customer,  the  systems  in  the  network  could be managed as
        either  a   single   facility   with   one   operational   and
        administrative  staff or as a set of individual systems linked
        together by the high speed network.

        There are  many  market  places  for  local  networking.   The
        following is a list of the most important markets.

        High End Systems

           There are a number of DECSYSTEM-20 customers who  currently
           own  the  largest system in the DECSYSTEM-20 family.  These
           customers  have  no  way  to  expand.   Local  high   speed
           networking  permits  these  installations  to easily expand
           their capacity without incurring extensive extra management
           or  operational  costs.   This  local  high  speed  network
To:  List                                                      Page 10
LOCAL HIGH SPEED NETWORKS


           solution  works  in  the  general  purpose   time   sharing
           environment  as  well  as in a shared data base environment
           without   requiring   any   reprogramming    of    existing
           applications.

        Schools or Businesses with a Limited Yearly Budget

           Consider the  small  schools  or  businesses  that  have  a
           limited  yearly  computer  budget.  These organizations can
           never justify purchasing  in  a  single  year  all  of  the
           computing power that they require.  However, they could buy
           a smaller member of the DECSYSTEM-20 family  in  the  first
           year  and  then incrementally expand their system by adding
           another small system each year thereafter.   This  approach
           is  also  valid  for  businesses  whose  computer  usage is
           predicted to start slowly and build up over several years.

        Computer Managers Faced with Decentralization

           There are many companies that are large enough to  purchase
           computers  to be located in different departments.  In many
           cases the manager of the central  computer  facility  would
           like  to  control  all  of  the  computing resources in the
           company  but  cannot  because  of  the  diversity  of   the
           requirements.   Local networking provides the facility that
           the manager needs  to  control  each  system  and  yet  not
           jeopardize  the  goals set for each individual system.  The
           leverage that could be gained by selling local networks  as
           a    means   for   retaining   centralized   control   over
           decentralized systems should be significant.





3.4  High Availability


        Another of the major benefits of the  local  network  is  high
        availability.   There are varying amounts of availability that
        can be achieved depending on the amount of hardware redundancy
        that  is  designed  into  the network.  At a minimum, the mere
        addition of the second system  into  the  network  raises  the
        availability  of  the  computing resources significantly.  The
        following  features  of  the  local   network   increase   its
        availability.

        Loosely Coupled Multiple Processors

           As systems are added into the network, the  probability  of
           at least one of these systems being up at any point in time
           rapidly approaches  1.0  (excluding  catastrophic  failures
           such  as  power  failures that affect all of the systems at
           once).  If one system is down, users can dial into another.
To:  List                                                      Page 11
LOCAL HIGH SPEED NETWORKS


           The  load  may  be  increased,  but  at  least the computer
           facility is still available.

           Since the systems in the local network are loosely  coupled
           (each  system  is  a  separate stand alone system connected
           only by a single cable), there  is  minimal  chance  either
           electrically  or  through  software  that  one  system  can
           adversely affect any other system.  This physical isolation
           greatly increases the overall reliability of the systems in
           the network.  It is also  this  architecture  that  permits
           large   numbers   of   processors  (as  well  as  different
           processors of the same family)  to  be  connected  together
           instead  of  just two as is common with the tightly coupled
           systems.

        Dual Ported Disks and Tapes

           If some of the systems in the network are physically  close
           to  each  other,  then it will be possible to dual port the
           disks and tapes between them.  In this case,  when  one  of
           these  systems  is  down,  its  disk and tape drives can be
           switched to the other system.  

           Since the disk file  structures  all  have  a  unique  name
           across  the  network,  the structures can be moved from one
           system to another without  requiring  any  modification  of
           user procedures.  This allows disk packs to easily be moved
           from a system that is down to another system in the network
           thereby providing high availability for the important files
           and data.

        Automatic Recovery on a Crash

           If a system in the network crashes, all of the users  using
           that  particular  system  will  have  to  log  in again and
           restart from their most recent  checkpoint.   However,  the
           users  who  were  logged into another system in the network
           and who were reading or writing files on a  disk  connected
           to  the  system that crashed, these users will not lose any
           of  the  I/O  that  was  in  progress.   The  monitor  will
           accomplish  this  by  reestablishing  the state of the open
           files after the crashed system is reloaded.  In the case of
           two systems whose disks are dual ported between each other,
           when  one   system   crashes,   the   other   system   will
           automatically  take  over the disks that were in use by the
           crashed system and will reestablish the state of each  file
           that  was open at the time of the crash.  In this case, the
           delay noticed by a user of files on that disk will only  be
           a few seconds instead of the normal system reload time.
To:  List                                                      Page 12
LOCAL HIGH SPEED NETWORKS


3.5  Performance


        Table 2  describes  the  disk  I/O  rates  that  are  commonly
        experienced  (or  predicted)  on  the  various  members of the
        DECSYSTEM-20 family.   The  average  rate  represents  what  a
        loaded  system  typically  sustains  over  a period of several
        hours.  The peek rate  might  be  reached  for  short  periods
        during  file  transfers  or  other  heavy  I/O activities.  It
        should be noted that these rates  include  both  swapping  and
        file I/O.  The rates in table 2 (and table 3) are specified in
        terms of pages per second.  A  page,  the  minimum  size  disk
        transfer  ever made by DECSYSTEM-20s, is made up of 512 36-bit
        words (4 128-word disk sectors).

                                 Average          Peek
                System          Pages/Sec       Pages/Sec
                ------          ---------       ---------

                Minnow           15 - 20         30 - 40
                2020             20 - 30         40 - 50
                2040             40 - 60         70 - 90
                2060            100 - 120       150 - 200
                Dolphin         250 - 300       350 - 450

                                 TABLE 2

        Table 3 describes the net data transfer rates  that  could  be
        achieved  over  various  high  speed network links.  The table
        also indicates the worst case  degradation  in  response  that
        could  result  due  to the added time required to transmit the
        data over the link.  These  worst  case  response  values  are
        given  as  a  percentage  of  the  response time that could be
        achieved on a system whose disks were all  locally  connected.
        The I/O transfer rates assume that the links are utilized at a
        60% efficiency level.  

                                                      Worst Case
                  Link Speed       Net Rate          Responsiveness
                  ----------       --------          -------------

                 1 Mega Baud      32.5 Pages/Sec         33 %
                 5 Mega Baud     162.8 Pages/Sec         71 %
                10 Mega Baud     325.5 Pages/Sec         82 %
                50 Mega Baud    1627.6 Pages/Sec         95 %

                          (Assumes 60% Efficiency)

                                 TABLE 3

        From these two tables it should be evident that a  link  of  5
        mega  baud  will  be  sufficient  to  support two 2060s linked
        together without undue degradation in throughput.  However, it
        would still be possible with some programs to incur noticeable
To:  List                                                      Page 13
LOCAL HIGH SPEED NETWORKS


        degradation in response time if all of the file  I/O  is  done
        over  the link.  In this case and in the case of more than two
        2060s linked together, the link should be closer  to  10  mega
        baud to achieve reasonable throughput and responsiveness.  The
        50  mega  baud  link  guarantees  that  the   throughput   and
        responsiveness  will  be  satisfactory  for  a  multiple  2060
        network.





3.6  Hardware Requirements


        ICCS Bus

           The HYDRA project is designing a  high  speed  inter-system
           bus called the Inter-Computer Communications Switch (ICCS).
           As currently planned, the ICCS bus will be a triaxial cable
           of  up  to  150  feet  in  length and it will be capable of
           running at a gross speed of 50 mega baud (with an  expected
           net  efficiency  of 60%).  The ICCS bus will support a star
           configuration and will allow up  to  16  components  to  be
           inter-connected.

           The ICCS bus is  definitely  a  requirement  for  providing
           large transparent multiprocessor configurations using 2060s
           or Dolphins.  For configurations that  can  be  set  up  to
           minimize  the I/O traffic across the inter-system link, the
           high speed ICCS bus would not be necessary.

        Ring Net

           Since the high speed ICCS bus  is  limited  to  very  short
           distances  (150  feet),  a  longer  system inter-connect is
           needed to satisfy the network configurations that  span  an
           entire  building  or  cluster  of  buildings.  However, the
           trade-off  for  longer  distances  is  slower  transmission
           speeds.   There  are  several possible approaches available
           for providing longer links.  MIT has developed an 8-10 mega
           baud link called "Ring Net" that is capable of being run up
           to 1000 feet between nodes (or between repeaters) for up to
           256  nodes.   This  design  has  one restriction in that it
           requires that the network be in the shape of a  ring.   DEC
           is  also  looking at building a slower speed version of the
           ICCS bus that would run at about 5 mega baud for  distances
           of 1000-2000 feet.

           These slower speed links would  be  well  matched  for  the
           smaller   members   of   the  DECSYSTEM-20  family  or  for
           configurations where the file I/O  being  performed  on  an
           individual system in the network is reasonably localized to
           that system.
To:  List                                                      Page 14
LOCAL HIGH SPEED NETWORKS


        MX20

           As a backup strategy, we could implement the local  network
           software  using  two  2060s  linked together by an MX20 and
           shared memory.  This approach would allow the local network
           software  to  be  developed  for  inclusion in Release 5 of
           TOPS-20 without out the risk of the ICCS bus  slipping  out
           of the Release 5 time frame.





3.7  Restrictions


        Shared Writable Pages

           There is one restriction with this local network  approach.
           It  will  no  longer  be  possible to use shared memory for
           process  synchronization  unless  the  processes  are   all
           running  on the same CPU.  Although there is no restriction
           on having shared writable memory, the processes using  this
           facility   must   use  ENQ/DEQ  and  UFPGS  to  avoid  race
           conditions.





4.0  20/VAX COMPATIBILITY


        This  section  discusses  how  the  DECSYSTEM-20   distributed
        processing  strategy  fits in with the corporate VAX strategy.
        It is very clear that DECSYSTEM-20 users will never be able to
        merely pick up their programs, bring them to a VAX system, and
        run them without  out  modification.   However,  there  are  a
        number  of  steps  that  we  are  taking to make it easier for
        DECSYSTEM-20 and VAX systems to coexist together.  

        DECnet 

           The DECSYSTEM-20 and VAX processes can already  communicate
           to  each  other  using DECnet task-to-task links.  With the
           future releases the DECnet networking capabilities will  be
           enhanced to support complex topologies and routing.

        File Transfer via DECnet and ANSI Tapes

           Release 4 of TOPS-20 will support  file  transfers  to  and
           from VAX systems using both DECnet and magnetic tapes.  The
           DECnet file transfer capability will be based on  the  Data
           Access Protocol (DAP).  Magnetic tape compatibility between
To:  List                                                      Page 15
20/VAX COMPATIBILITY


           the DECSYSTEM-20 and VAX will be based  on  the  ANSI  Tape
           Label Standard.

        Inter-system Job Submission

           Release 5 of  TOPS-20  will  allow  DECSYSTEM-20  users  to
           submit  jobs  over  a  DECnet  link  to  a  VAX  system for
           execution.  This facility will be implemented using the DAP
           protocols.

        Network Virtual Terminals

           Perhaps the most important  part  of  20/VAX  compatibility
           will  be  the  network  virtual  terminal  facility.   This
           mechanism will allow DECSYSTEM-20 users to set  their  host
           to  be  the  VAX system.  They will then be able to use the
           VAX system as if they were directly connected  to  it.   By
           using  a  combination  of  the  file transfer facility, the
           inter-system  job  submission  facility,  and  the  network
           virtual  terminal facility, DECSYSTEM-20 users will be able
           to gradually start converting old applications and  writing
           new applications for the VAX system while continuing to run
           their "bread and butter" production  DECSYSTEM-20  programs
           without perturbation.

           It should be noted that the current VAX plans only  address
           being able to support network virtual terminals between VAX
           systems.  However, the DECSYSTEM-20  will  support  linking
           terminals  to  VAX  systems as well as other DECSYSTEM-20s.
           This means that users who need to  switch  back  and  forth
           between  the  two  systems must be using a terminal that is
           physically connected to the DECSYSTEM-20.

        Network Mail

           A  corporate  network  mail  protocol  is  currently  being
           developed.  This will allow both VAX and DECSYSTEM-20 users
           to correspond between their systems.

        Data Interchange Utility

           A  utility  to  interchange  data  between  DECSYSTEM-10/20
           systems  and  VAX systems is currently being designed.  The
           goal of this project is to allow easy  conversion  of  data
           files from the 36-bit format to the 32-bit format.
To:  List                                                      Page 16
20/VAX COMPATIBILITY


4.1  20/VAX Growth Strategy


        As the corporate VAX  strategy  starts  to  take  effect,  the
        number of VAX systems in the field will increase dramatically.
        Likewise, the range and flexibility of  the  VAX  family  will
        also  grow.   It  will  become  more  and  more  desirable for
        DECSYSTEM-20 customers to coexist with VAX systems.  The tools
        described  above  provide the basis for this coexistence.  For
        example, consider a DECSYSTEM-20 installation faced  with  the
        problem   of   expansion   in   1985.    By   that   time  the
        price/performance  offerings  of  the  VAX  family  will  have
        surpassed  those of the DECSYSTEM-20 family.  The installation
        may decide that it is time  to  expand  by  purchasing  a  VAX
        system  and to put all new applications on the VAX system.  In
        the mean time, the older production jobs will continue  to  be
        run  on  the DECSYSTEM-20.  The VAX system would be brought in
        and linked to the  DECSYSTEM-20  with  one  of  the  supported
        inter-connect  strategies  described  previously.   The use of
        network  virtual  terminals,  file  transfer,  and  the   Data
        Interchange  Utility  would  then  permit  an  easy transition
        between  the  two  systems.    This   transition   should   be
        considerably  less  traumatic  than  switching  to  any  other
        vendor.