Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site fortune.UUCP
Path: utzoo!watmath!clyde!burl!we13!ihnp4!fortune!phipps
From: phipps@fortune.UUCP (Clay Phipps)
Newsgroups: net.arch
Subject: IBM "Single Level Store" / Re: Why Not Virtual Files?
Message-ID: <3179@fortune.UUCP>
Date: Fri, 27-Apr-84 18:15:04 EST
Article-I.D.: fortune.3179
Posted: Fri Apr 27 18:15:04 1984
Date-Received: Sat, 28-Apr-84 10:26:15 EST
References: <333@oliveb.UUCP>
Organization: Fortune Systems, Redwood City, CA
Lines: 45

----------

The long-awaited-but-cancelled IBM Future Systems (FS) line of mainframes
supposedly incorporated the concept of "single-level store", essentially 
what the original posting described.  It reportedly used 64 bits of address.
It was to be a "capability machine".
I am personally convinced that the "Mass Storage System" ("Son of Data Cell")
was designed for FS, but later grafted onto the S/370 line when FS was canned.
MSS takes the place of tape archives, using a LRU algorithm (flawed in MVS)
to select files to migrate to MSS, and demand-restoring absent disk files
to disk when they are referenced.

The IBM System/38 implemented many of the rumored FS concepts.
It has a "single level store" that uses a 48-bit address,
which gave every one of the 281 trillion bytes of data on the system,
regardless of where it was stored, a unique address.
The addressing scheme uses hash tables to avoid having the 1 billion bytes
of resident page tables that the S/38 would conventionally require 
for addressing.  I doubt that this scheme is very speedy, however.
One description of this is in

    Merle E. Houdek and Glen R. Mitchell (both of IBM GSD):
    "Hash Index Helps Manage Large Virtual Memory",
    *Electronics*, March 15 1979, p. 111 ..113.

IBM also published a collection of papers under a name something like
"System/38 Technical Description" during the same time period.

I believe that the British (English Electric ? ICL ? U of Manchester ?) Atlas
had a "single level store", and was (if my memory serves) the first computer
with paged virtual memory, back around 1960.  The file and memory system 
was fairly simple, with paging used primarily to avoid memory fragmentation,
so I'm not sure that it matches your description.
It's described in the 1st edition of Bell & Newell [which I don't have
available right now, naturally].

Of course Burroughs has probably done work in this area, too.
I believe that they had logically independently segmented memory 
before anyone else.

-- Clay Phipps

-- 
   {cbosgd decvax!decwrl!amd70 harpo hplabs!hpda ihnp4 sri-unix ucbvax!amd70}
   !fortune!phipps