Persistence and Newton OS...

Fare Rideau
Fri, 4 Apr 1997 15:50:42 +0200 (MET DST)

> The reason we're not considering that at the moment is that it's totally out
> of reach.  The memory needed for that is SRAM or low-power DRAM, which is
> either enormously expensive or immensely slow.
Actually, we *are* seriously considering persistence,
but in a software, not hardware, implementation.

   The idea is that "main memory" DRAM should be considered
as just another cache between computations and persistent storage on disk,
much like registers and various levels of SRAMs before.
The fact that DRAM be a software managed cache, not a hardware-managed one,
is just irrelevant to the user/programmer
(but the ones who do implement persistence;
but there are implementers of hardware caches, too, so that's no big deal).

   Of course, we do not guarantee *immediate* flushing of the DRAM "cache"
into disk (or network store), so that to be resiliant to power shutdowns,
we must ensure that at any time, a *consistent* version of the system
is recoverable from disk, no older than some hardware-performance-dependent
time (typically, from a few seconds to a few minutes).

   See Eumel for a system that has provided persistence on PCs
used in the industry for more than 15 years...
See Grasshopper OS, Texas Persistent Store, and more,
for systems with lots of docs about how to achieve it.
[All that in the Review/ subproject].

> A far better method is to manage storage so that there's no
> application-dectectable difference between persistent objects and temporary
> ones, and then placing the persistent object in persistent storage.
> Actually, the Newton does some of that, but it's not particularly good at
> it.
Indeed, that's software-implemented persistence.
Again, that persistence be implemented in software or hardware
is conceptually no more than multiplies, pattern-matching,
and compiling being done in software.
Surely, all of these could be done in hardware,
but software makes everything cheaper, more portable, and more flexible.

> If the Tunes pages don't discuss NewtOS it's probably because of lack of
> time -- it does do some things well, and should be researched.  I prefer
> Geos, but then everyone here knows that.  :)
There are a few pointers to the Newton OS on the OS review page,
but I've not read all the papers presented
(the few I read were mostly commercial hype,
and/or not technically very interesting).
   Feel free to contribute comments, reviews, etc.
== Fare' -- -- Franc,ois-Rene' Rideau -- DDa(.ng-Vu~ Ba^n ==
Join the TUNES project for a computing system based on computing freedom !
                TUNES is a Useful, Not Expedient System
URL: ""