I hate to open a can of worms here, but feel I must.... However, I am not
a poacher myself, nor do I advocate it. I only write this to tell you why
I don't feel sorry for folks who queue up WAY too much work and then gripe
about it when someone else calls them on the carpet about it by poaching
them. I write this in the hopes that you'll see the error of your ways,
and work not just for yourself, but for the good of the group.
My conclusion at the end of this message is for George's consideration, and
the rest of this message defends this conclusion:
George: v20, *and* the PrimeNet server, ought not allow any one machine to
keep more than ten times its average communication frequency in exponents
queued, and no more than 60 days no matter what -- requests for additional
exponents when the server knows that that machine already has two months'
work ought to be denied. If a machine reports in every 3 days, let it
keep no more than 30 days. If it reports in daily, let it keep ten
days. By stopping exponent hogs from locking up hundreds of exponents
just because they like the small ones, GIMPS will reach its goals
(milestones, proving M37, etc), much faster.
----------------------------------------------
Dave has at least 80 exponents reserved between 2.4M and
3.99M. Eighty. Almost all are less than suspected M37. It is a
certainty that without poaching, we will have to wait until late 2000 or
later to prove M37, because Dave is trying to do all the double-checking
singlehandedly.
I cannot stress this part enough: This is why we have thousands of
participants in GIMPS! It is our PRIMARY raison d'etre! To spread
around the workload to get things done faster! By trying to take 80 of
the 280 or so exponents left for doublechecking up to M37 (nearly 30% to
ONE participant!), Dave is intentionally thwarting the very purpose of
GIMPS: distributed mathematical research. DISTRIBUTED computing is key!
Each of Dave's 80 exponents will take a P-II/400 0.09 seconds per
iteration. If the average exponent is closer to 2.8M, here's how much time
Dave has set aside:
2.8M x 80 x 0.09 = 20,160,000 seconds DIV 86,400 = 233 days.
That's if Dave uses P-II 400's, on PRIMARY tests. Double-checks use a
different LL code, and take longer. I doubt Dave is using P-II's for this
purpose, too. If it's P-90's, that's 233 x 4.5 (times slower) x 1.2
(times slower to double check, a guess), or 1574 days of work queued up for
Dave.
I can only find six named machines of Dave's in the work list. 1574 days
of work over 6 machines is an AVERAGE of 262 days of work queued up per
machine.
What a PIG. Why does ANYONE need nine months of work queued up,
especially for machines that seem to report back to GIMPS on a daily
basis? Many of us want to see results -- we want milestones, we want to
see "All exponents less than 3,000,000 have been double checked." We want
to see "Double checking proves 3021377 is the 37th Mersenne Prime".
Most of Dave's assignments have gone untouched for 30 - 90 days.
We don't want to wait a YEAR for this milestone, just because you and a
handful of others want to test all the little exponents.
Your machines are useful to us, don't get me wrong. Nobody here wants you
and Dave (and other exponent hogs) to quit GIMPS. We just want you to
reserve a reasonable number of exponents, and take what comes to
you. These machines will be equally useful to us whether double-checking
2916117 or 4717123.... and we'll get where we're going faster that way!
Dave's machines are permanently connected (or frequently connected) -- they
have reported progress nearly daily -- slow, steady progress, but they
report.
Thus, IMO, Dave should not have his clients set to queue up more than TWO
DAYS of work. I set mine at ONE day, so that I don't even get a new
assignment until the machine is less than a day away from finishing its
exponent and being left with NO work. And that's the way it ought to be
-- nobody ought to even be ABLE to hold up the progress of the group in
reaching milestones for this long. When your machine is ALMOST out of
work, THAT is the time to request the smallest available exponent OF THAT
MOMENT.
So, to your paragraph below, there's nothing wrong with seeking out the
smallest available exponents.... but there *is* something wrong with
seeking out nine months' worth of them, and holding up the very purpose of
the group.
If Dave gets poached, I won't shed a tear.
I'd have done a similar analysis on your assignments, but didn't know your
ID. You're probably not as heinous as Dave is, since he appears to be the
worst of the lot on cursory inspection, but ANYONE holding more work than
necessary is on the list of "won't cry for you, Argentina" folks.
MOST folks understand this. There are 26,600 machines right now, and
44,200 exponents assigned -- 1.66 exponents assigned per machine. Since
the software defaults to 28 days of queued work, this is
understandable. Dozens per machine is just not defensible, under any
circumstances (except, perhaps, multi-processor machines, but then ID's
would be different).
Any defense you'd like to offer for holding 9 months' work, I'll listen to,
but I doubt you'll come up with anything convincing.
At 07:42 PM 2/3/00 -0900, you wrote:
>Not a big deal in the greater scheme of things, but frustrating to people
>like diamonddave and myself who make an effort to seek out the smaller
>exponents and reserve them.
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
