Aaron Blosser writes:
   I don't suppose George could just program something into the code
   to have it check for the user being idle (like the screen saver
   check does, but independent of the system screen saver routines)
   such that if the user doesn't hit a key or move the mouse for xx
   minutes, it would begin it's calculations (still at whatever
   priority you set it to...idle by default), but when the user is
   hitting keys or moving the mouse, it'll stop calculations
   altogether?  That may allay the (unfounded) fears of some that
   Prime95 somehow steals cycles from other running programs.

Unfortunately, I have personal experience in this area, though not
with Prime95.  My own (UNIX-based) Mersenne programs and scripts, from
before GIMPS started, included checks not only that all logged on
users were idle for at least three hours, including their terminal,
mouse, and keyboard, but also that the load was only the 1.0 due to
the program itself (and less than 0.1 or so when starting).  These
checks themselves (that the load remained low and any users were still
idle), when performed every two minutes under SunOS 4.x on the
SPARCstation 1's and 2's that were available at the time, usually
kicked the load average up another 0.5 or so.

But two minutes is quite a while to wait for something hogging your
CPU to stop.  At least according to the ten or so people that
complained out of the roughly 100 computers my scripts were running on
for a couple of years.  And the scripts were careful to start only
after hours, even if the computer appeared idle during the day.  And
the programs that did the actual work (almost always trial factoring
because I didn't have an FFT-based LL program) always ran at the
absolute lowest priority UNIX offers.

Note further that checking to start things was done remotely; there
was _no_ process of mine on the local machine when it was not idle,
not merely a process only checking for idleness: the load average and
user list could be checked without any local process.

There were _still_ complaints.  Even though the only thing that some
of them could point at that indicated "slowness" was the load average
being 1.0 instead of 0.0.

So, no matter how much CPU you think this sort of change could gain
GIMPS, I must suggest that it _not_ be done.

Except - perhaps - under the control of another .ini variable and the
default is to do things the current way.

                                                        Will
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to