Re: ????MicroShield Validation; Physical Testing

[Gary Masters <gmasters@csn.net>]

Jim Barnes,

We don't use MicroShield, but, I question whether someone else's
validation would be useful to you. Our practice is that any safety related
code must be validated on each platform on which it is installed.  I also
think this is eluded to in DOE order. 

The primary reason for this practice is illustrated by a report in an
Operating Experience Weekly Summary from 1994. In this report some folks
at PNL detected a floating point math error in some PCs (not Pentiums)
caused by a ROM problem. 

>>>>>Begin OEWS article<<<<<

Operating Experience Weekly Summary 94-1

December 31, 1993 through January 6, 1994

3.      COMPUTATIONAL PROBLEM WITH MATH COPROCESSOR IDENTIFIED ON SOME
PERSONAL COMPUTERS



On December 17, 1993, personnel at Hanford Pacific Northwest
Laboratory reported that a potential computational problem
exists with some IBM and IBM-compatible personal computers that
have Intel math coprocessors.  Laboratory personnel determined
that some personal computers return erroneous results from
floating point calculations using a "divide-by-zero" operation,
and they advise that calculations performed on computers
exhibiting this problem be checked for accuracy.  (ORPS Report
RL--PNL-PNLBOPER-1993-0043)



Following is a list of personal computer configurations
identified by Pacific Northwest Laboratory that exhibited this
problem.



  - Zeos+ 486DX with Award+ Bios v3.1, revisions
    452-0005-02A and 452-0005-01B



  - Gateway+ 486DX with Phoenix+ Bios 0.10 G14



  - IBM+ 486DX Valupoint+, Bios not identified



  - Several unlabeled systems with a 386DX cpu, a 80837
    coprocessor, and an AMI+ Bios installed



Laboratory personnel believe the problem is related to improper
initialization of the math coprocessor when the computer is
initially turned on.  They also stated that they have not
determined that all systems with the listed configurations have
the problem, only that some have exhibited the problem.  When
contacted by investigators from Pacific Northwest Laboratory,
the vendors involved stated that this was the first indication
they had of the problem.  The vendors were able to reproduce
the problem and are evaluating it to determine if a fix can be
implemented.



In the interim, personnel at Pacific Northwest Laboratory
developed two test programs called FPTEST.EXE and DIVZERO.EXE
to test for proper math coprocessor operation on an IBM or
IBM-compatible personal computer.  These two programs and a
README.TXT file are available to DOE personnel through Internet
via ftp to ftp.pnl.gov and are found in
directory\pub\outgoing.  At sites without Internet access, DOE
personnel can contact the Pacific Northwest Laboratory Customer
Service Center at 509-375-6789 for assistance.



DOE personnel are encouraged to test their personal computers
for proper math coprocessor operation.  If erroneous results
are obtained during the test, they should independently verify
the accuracy of safety-related calculations performed on that
particular personal computer.

>>>>>End OEWS article<<<<<


  *********************************************************************
  * Gary Masters                                 Phone (303) 966-3266 *
  * Radiological Engineering                       Fax (303) 966-8459 *
  * Rocky Flats Environmental Technology Site                         *
  * PO Box 464   Building T690A                Email gmasters@csn.net *
  * Golden, Colorado 80402-0464                                       *
  *********************************************************************