Shift Happens

[Paul Oakes <POakes@TECHDEV.CompuServe.com>]

Date:	Wed, 1 Apr 1998 -8:00:00 -0800
From:	Douglas Moran <moran@ai.sri.com>
Subject:	Funding for a new software paradigm

(Washington, DC, press release by IP Newswire, 1 April 1998) The 
Defense Advanced Research Projects Agency (DARPA) today announced a 
major new initiative in software engineering.  F.P. Rivers, program 
manager for the initiative, said that it addresses a major problem 
facing the US military: that much of current information technology is 
too "compute-intensive" to be deployed where it is most needed-at the 
small unit or even individual soldier level.
This initiative has its origins in a fortuitous observation: Rivers 
and several colleagues noticed that users on the most widely used 
platform-
Windows 95 -- were routinely presented with messages that an unknown 
unrecoverable error had occurred, and that these users just as 
routinely ignored those messages.  "This occurred not just in casual 
use, but also in mission-critical operations."
Rivers said, "Once we started thinking about these messages not as a 
help, but as a hindrance, several other observations came together." 
 In a typical program, 40% to 80% of the code is devoted to error 
detection and error handling.  "Software bloat"-the ever increasing 
size of programs-has been blamed on programmers adding more and more 
features, but could also be blamed on all the error handling 
associated with those features.  To make matters worse, multiple 
studies had shown that much, if not most, of the error-handling code 
was never tested.  Sometimes this was because of time and budget 
pressures; sometimes the potential errors were so obscure and complex 
that the situations were too difficult to create "in the lab".  This 
research was backed up by actual experience: error-handling code was 
often found to have significant errors.
Rivers summarized, "So, the typical program is overloaded with code 
that is rarely used, that may not work, and whose output is likely to 
be ignored anyway."  He concluded, "With this code removed, programs 
will be dramatically smaller and will run somewhat-to-noticeably 
faster."
Many software developers, including several major vendors, have 
already taken some tentative steps in this direction, having 
recognized pieces of the problem, but without grasping the "big 
picture".  Rivers said he expects this new approach, dubbed 
"Fault-Oblivious Computing", to quickly become the dominant 
software-engineering paradigm.  He acknowledged that there were small 
highly specialized segments where fault-tolerant computing and program 
verification would still be of value.  A major component of this 
initiative will be to develop tools to automatically identify and 
remove unneeded error-handling code from existing applications.
The success of this approach would be be bad news for memory-chip 
manufacturers, who are already hard-hit by decreased demand.
[Perhaps Fault-Oblivious Computing could be used to help with the Y2K 
problem, getting rid of all those gratuitous date comparisons!!! 
 PGN]


RISKS Digest 19.64