Newsgroup: alt.best.of.internet

This appeared in comp.unix.bsd.  Bill Jolitz is the person who put
386 BSD together.


Diomidis

vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv

Newsgroups: comp.unix.bsd,comp.os.386bsd.development
From: jmonroy@netcom.com (Jesus Monroy Jr)
Subject: [386bsd] Error Theory
Message-ID: <1993Feb17.091158.27775@netcom.com>
Keywords: Error theory device driver computer theory
Organization: Netcom - Online Communication Services  (408 241-9760 guest) 
Date: Wed, 17 Feb 1993 09:11:58 GMT
Lines: 135


The following is message I sent to William Jolitz a while back.

---------------------------------------
======================================= 

   
 
 
        RE: Error Theory/Error bit
 
 
        Dear Bill,                              03:10:44 Mon  01-04-1993
 
 
                In my most recent efforts to rewrite the FDC code in use
        for 386BSD, I have been unable to find a correct paradigim in
        which recovery for an insatanciated error can be salvaged in a
        concise manner. Before giving you my thoughs I must share some
        fundamental Ideas I think that we can agree on.
 
 
                1) At present we measure time as decay.
 
                   ie. decay of the the sun rising, a clock ticking
                       the half life of Uranimum
 
 
                2) An Error is and embedded or integral process.
                   An Error can be removed from a process.
 
                3) An Error can cause failure.
 
                4) A Failure is not an Error, it is a response to an
                   Error.
 
 
                My thoughts have been that the premise for my algorithm
        is incorrect. That is most code is written with the thought
        that the code will run correctly most all of the time, this
        part is correct for efficient and active code, but treatment
        of errors is limited at the present to traps. Traps being an
        enclosure or encapsulation of the data recovery process, with
        most if not all data being information that the unit must
        digest.
                Because the Error Process is feed or contains data
        extranious to the Instanciated Error, a recovery calls for
        sifting through data that is completely unassociated with the
        current. Current being the time flow which speaks to the process
        or error (not either or both for the process and the error are
        integrated irregardless of the discrete state of the whole
        process.).
                In my recent readings I have found that the Egyptians
        may have considered "Erra" a way of life.  Carl B. Boyer writes
        in "A History of Mathematics", that the Ahmes Papyrus considered
        the unit fraction 3/5 to be 1/3+1/5+1/15. They used a method of
        derivation that used the "natural fractions", namely 1/2,1/3 &
        2/3.  Boyer and others drew no conclusion from these inferences,
        but it is somewhat obvious that they have taken the route of
        "least reflected" errors. This is easily seen with the example
        given.   With no precision method  of calibration, and we and
        they knew this to be true, they resorted to a method which could
        easily be explained and logically be derived.
                3/5 is a number, which if taken from a whole item (such
        as an apple) is difficult to conceive. If on the other hand I
        said  give yourself  1/2+1/5+1/15 you might have a better
        estimate.  You may say, "Well, What if I cut 1/5 and triple
        that?".   Consider the error!  Your magnitude of error is now the
        larger part of the error times 3.  If, however, you use
        1/2+1/5+1/15, your magnitude of error is reduced in sequence by
        the magnitude of the partial size of the parts, respectively.
        That is the magnitude of your error is no larger than:
 
 
                        n(A + 2/3A + 4/9A ...)
 
                or      n(A + 1/2A + 1/4A ...)
 
                or      n(A + 1/3A + 1/6A ...)
 
 
                        VS.
 
 
                        n(A + A + A ...)
 
        if you cut your apple in 1/5 parts.
 
                This being the case "Erra" can be seen as something
        always known. Definitely we are told Adam and Eve created the
        First Error of/for mankind.
 
                When I preposed this idea to an associate, he said it
        had been done and what I might conceive would be a waste of
        time. Further, he stated that since it had been done I would,
        with all probability have nothing new to add or insight.  I had
        to point out that like music theroy, the Ideas came first,
        theory is a description of It.
                It sounds a bit ludicrous that one might consider what I
        prepose as a science, but in fact if the development is not
        considered, my personal feeling is that computational science
        will not evolve as we dream/hope.
                A strong point in my favor is that computer science in
        most levels of higher education is an extension of the
        Electronic Engineering Department, when in fact the software is
        really exclusively a branch of logic and error decection.
        Another major factor is the misnomer that software development
        is always behind hardware development. This is altogther
        incorrect, we know what we want to do, we are confinded by EE's
        with dillusions of being the next Shocky.
                Aristotle and the Greeks called it "Logos eritakos",
        in English, quibble. We might call it "Eritalogics".
 
                How can this be implemented in 386BSD you ask?(maybe)
 
                I am not quite sure, but I can see major parts in my
        mind, not the whole.
 
 
                Will this require major hardware redesign?
 
                No, but to work in all efficency the CPU must have a
        Hardware & Software ERROR Bit. In the IBM BIOS, that runs most
        ATs, the system service call rely heavily on the "carry bit" to
        double as an "error bit".  This tells us that real time single
        operator Operating Systems have recognition of what we need to do.
                I have other thoughts about 386BSD they will follow in
        other letters.
 
 
                                Jesse Monroy Jr.
                                jmonroy@netcom.com
 
 

-- 
Diomidis Spinellis    Internet: <dds@doc.ic.ac.uk>  UUCP: ...!uknet!icdoc!dds
Department of Computing, Imperial College, London SW7     #include "/dev/tty"

Unless otherwise expressly stated, all original material on this page created by Diomidis Spinellis is licensed under a Creative Commons Attribution-Share Alike 3.0 Greece License.