[comp.os.vms] Suggestions for Large/Fast Computers from DEC

rlb@rtpark.rtp.ge.COM (Bob Boyd,8*565-3627 24-Jun-1987 0153) (06/24/87)

  Terms(glossary)

File System     hardware+software defining the file system

System          a single security domain

CPU             a processing entity (compute engine)

Processor       a component of a CPU

Channel         communication path/facility

Storage         media  and  drives  providing  bulk  data  storage  on-line  or
                off-line

Memory          CPU resident data storage

  With respect to "Large" computer systems The following topics seem to  me  to
be important:

   - A  file  system  which  provides  high  speed  access  to  most often
     requested information about many files --  possibly  by  providing  a
     larger  quantity  of  information  about  files  in  the  directories
     themselves in addition to or instead of the file headers.   Eliminate
     requirements  to "open" or "connect" to a file to get size, security,
     dates, and other selected information.

   - I/O structures to accomodate a high rate of transactions.   Areas  to
     investigate include multi-arm disk drives, file system co-processors,
     higher speed "channels" from disk to  processor(s),  caching  at  the
     controller  level,  configurable  semi-automatic  or  automatic  disk
     optimization,  parallel  multi-channel  data   paths   to   eliminate
     contention.    Focus  on file system(s) and architecture(s) and allow
     for 3rd party contributions to storage technologies.

   - Provide very high speed inter-CPU communication channels,  especially
     important for large, fast, cluster-like systems.

   - Accomodate  mixed  asynchronous  &  block mode terminal I/O with very
     intelligent terminals/front ends -- implement more screen controls at
     the  front  end  --  field  attribute  checking,  refresh,  etc.  The
     objective here is provide the ability to free up  the  main  cpu  for
     more  computationally intense tasks.  (This sounds strangely like IBM
     327x style terminals! Did I really say this?)

   - Reduce the incremental costs of  adding  disk  storage  in  terms  of
     associated controller costs as the size of the system grows.

   - Provide   high-speed,   high-density,  easy  to  use  backup/archival
     system/media which  ties  into  DEC  or  3rd  party  offline  storage
     management  system(s).    The  system  and  media  must be capable of
     retaining material for at least 10 years, which  means  changes  must
     provide   compatibility   and/or   upgrade  of  database(s)  so  that
     backups/archives from years gone by are  still  manageable  with  new
     version/product.

   - Provide T1+ network interconnect capabilities.

   - Make  it  possible  for  the user submitting "batch" jobs to tell the
     system something about how much resource is to be used and  how  fast
     they  want  it  back  as  an alternative to precise controls a la VMS
     SUBMIT command.

   - Provide batch stream dependency networking (even  as  far  as  across
     system boudaries)

   - Separate  dump,  swap,  and  page  files from boot device(s).  Larger
     disks are not  sufficient  to  improve  the  manageability  of  these
     resources.  Provide configuration controls to specify more about disk
     configuration at boot time.

   - Provide configurable scheduling algorithms providing a mix of  class,
     resource,  capability  based  and  other  algorithms.   With a mix of
     multiprocessing and uni-processing  this  becomes  critical.    Large
     systems  will  be  required  by  user  community  to  provide certain
     "guaranteed" levels of service.

   - Provide change  control  and  notification  capabilities  for  system
     software and hardware for online checking and regression testing.

   - Provide  system  capabilities  for  coping well with disk failures --
     e.g. disk drive with a paging file open croaks -- after timeout  kill
     or  attempt  to  re-map  processes to other paging file(s) and finish
     dismount of disk.

   - Provide   callable   interface   to   environmental   monito(s)    --
     particularyly  temperature,  humidity,  1  external power tap per cpu
     cabinet.  Provide a skeleton environmental  monitor  program.    This
     will  enable  sites  to  do  observe  their  own  policies/quirks for
     deciding when to batten down hatches and try to save the ship.

   - Provide a configurable security monitor -- back end to security alarm
     facilities.

   - Provide configurable resource requirement and conflict scheduling and
     notification  capabilities.    This  capability  is  tapped  by   the
     following:    Develop  human  engineered  operator and system manager
     interfaces  for  interrogating  and  controlling  resources  and  the
     system.

   - Eliminate  requirements  for  standalone  backup  --  a capability is
     needed for backup of files held open by the system.

   - Provide  "directory  service(s)"   for   inter-user   communication--
     especially important for large corporate networks.

 Bob Boyd                     Usenet:    rlb@rtpark.rtp.ge.com
 DECUS VAX SIG                DECUS Large Systems SIG
 Commercial W/G Chair         Special Project Coordinator

 GE Microelectronics Ctr.     Voice:     (919)549-3627
 POB 13049, MS 7T3-01         GE DIALCOMM:  8*565-3627
 RTP, NC 27709-3049           GE DECnet: RTPARK::RLB
-----------------------------------------------------------------
Some of these ideas might be my own, and then some of them might
be yours.  And if they make you  $1 Billion then I ask that you 
promptly forward me 0.1% of your profit -- I'll take a personal
Cray equivalent system as a substitute.
----------------------------------------------------------------