Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!husc6!think!ames!oliveb!sun!rainier!dchen
From: dchen%rainier@Sun.COM (David Chenevert)
Newsgroups: comp.arch
Subject: Re: Size of SysV "block" (really: byte != 8 bits)
Message-ID: <23801@sun.uucp>
Date: Sun, 19-Jul-87 22:09:12 EDT
Article-I.D.: sun.23801
Posted: Sun Jul 19 22:09:12 1987
Date-Received: Mon, 20-Jul-87 03:49:58 EDT
References: <218@astra.necisa.oz> <142700010@tiger.UUCP> <2792@phri.UUCP> <933@geac.UUCP>
Sender: news@sun.uucp
Lines: 40
Summary: Bit explosion

In article <933@geac.UUCP>, daveb@geac.UUCP (Dave Brown) writes:
>   One of the IBM techies recently admitted in a public place that they're
> using up about 1 address bit per 18 months.  Ie, applications are getting
> larger rather quickly and they're having to scurry to keep up.
>   Multics suffered from only (!) having 36 bits worth of word, limiting
> segments (file) sizes significantly.  In fact, they promptly kludged up
> a file which was really a directory of subfiles.
>   Lisp and Prolog machines typically need extra bits for tagging data with
> its (primitive?) type.
> 
>   Expect *ANOTHER* upheaval when you have to convert your applications
> to a long word-length (or at least long address-length) machine.  One
> existing machine already uses 48 bits for a "C" character pointer (I
> won't mortify the manufacturer by mentioning his name).

  I'm curious about Multics file size being limited by 2^36 =
  64 billion bytes? Were those guys using triple-sided disks or what?

  On the general subject, I agree 100%.  My figures have address bits
  coming at more like every 14 or 15 months, but the basic point is the
  same.  You can get 128 Meg (27 bits) with your Sun-4, so we have
  about 5 * 1.5 years or so till we first barely touch the ceiling.  A few more
  years for, say, the Macintosh crowd, but still, by the end of the
  century, we're going to be scrounging for bits.

  Responsible/smart computer trendsetters (CPU manufacturers,
  operating system/ language people) will start defining 64-bit ints,
  and using them where cost is not important (e.g., the field which
  holds the length of a file).

  Expect to see an even wider variety of hacked/kludged/painful
  pseudo-solutions than on the 16 to 32 bit transition.  Intel will
  have a segmented 586 or 686, Dec will just have a VAX compatability
  mode bit on their VAVX (Virtual Architecture Very Xtended).  IBM
  will switch to floating point addresses.

  In little or no seriousness, I'd like to hear some other projected
  painful solutions to this problem.  Mail to me if you don't want
  them posted, or to the appropriate manufacturer if you think
  they'll listen.