Stratcat wrote:

"f/f george" <george@yourplace.com> wrote in message
news:ehl6h093d9vqgkcr6obl7v0vckp5sgfl62@4ax.com...

On Fri, 6 Aug 2004 18:11:13 +1200, "~misfit~"
<misfit61nz@yahoo-mung.co.nz> wrote:

There must be a few WUs getting lost out there. You know how each
unit is sent out with a _0, _1 and a _2 suffix so that each unit is
calculated three times and cross-checked? Recently I've noticed a
few _3's and _4's and today I got a _5. Obviously a few results are
being lost or not returned. My understanding is that, if all three
'units' haven't been returned by the expiry date they send out
however many other copies are needed to get the three results. A _5
must mean that all of the original three units were lost.

Just thought I'd mention it. Anyone seen one higher than _5? Just

curious.

It could also mean that the first returns weren't within
"computational" sameness. Meaning that the first return may have
asked for 35 credits while the second was asking for  68 credits and
the third was asking for  2 credits. Something is wrong and the units
would be sent out again and again until 3 returns are within alowable
limits. Those limits are set by the Scientists and can of course be
changed.

FWIW - I've gotten #'s all over the place, lately. 0's - 5's, with
many 3's & 4's, maybe even enough 3's & 4's to say a solid 25 - 30%
of 'em.

Erronious returns, or lost/missing returns? Take your pick. Only
Berzerkly

<Grin>. There should be an 'e' between the k and l, Berzerkeley, like that.

knows for sure. I'd bet more than a few peeps didn't return
on time due
to the ongoing probs, inadvertantly setting long cache estimates,
or simply bailing out. Prolly the last two are a definite minority,
but still a significant enough group to make a dent in the stats.

I know of a few impatient peeps who bailed.

My cache set to 6.2 days gives me 10 - 11 days of work. I've had to
reduce it to 5.2. I've seen peeps in the BOINC forums setting cache
to 12 - 14 days! Even if the client or scheduler are smart enough to
not give over 14 days work, it just takes a bit of a bad estimate, a
little machine down time, or some comm probs to start a cascade
causing the cached WU's to go stale.

Yep, I've just warned a guy in my team who had set his cache to 12 - 14
days. The TC (To Completetion) calculation isn't very accurate (as evidenced
by your paragraph above). I have one machine that has a 3:09 TC that
actually takes 2:45 (that's Ok) and another that has a 3:55 TC that actually
takes 5:30. That machine could be a problem with the cache set high. As it
is I have a 6-8 day setting on that machine's preferences (work).

That second machine mentioned is used for playing mp3's all day, and
sometimes for DivX's as well. The TC and cache settings assume that the
machine in question will be running 24/7, soley running BOINC as, when
benchmaking, the software stops computation and network activity. (If there
is any computation, other than BOINC, going on as BOINC usually benchmarks
at install or at start-up).

To me this system makes two fundamentally erronous assumptions. The first is
that the machine in question is *only* going to be running BOINC (it
suspends computation to benchmark) and sets the TC accordingly. The second
assumption, when filling the cache, is that the machine is going to be
running 24/7. Ok, for a lot of us here, that second assumption is correct,
but for a lot of others who will be installing it, it isn't.

Considering both these factors, and the fact that WUs are stale after 14
days, I wonder why Berkeley hasn't set a maximum cache size at seven days. I
mean, not everybody who installs BOINC reads the forums or this NG. I
haven't seen mention on the download or instruction pages of the fact that
the WUs go stale. (I might have missed it). It wouldn't be hard to have a
maximum setting on the preferences pages. As it is you can put 30 days there
AFAIK and get the WUs delivered to your machine. (I haven't tested this for
obvious reasons).

I guess in that light it's not surprising that some of us have WUs with a _8
suffix.

As always, my opinion is free. You get exactly what you pay for.

I'm happy with the transaction. <g>.
--
~misfit~