I have recently noticed that, since the credit-granting problems of a couple months ago, my RAC, after being bumped up to something amazing, has reduced down to zero and stayed there, despite trickles still being granted credit. Has anyone else had this happen? Is it a weird side effect of the trickling that CPDN does rather than just completing a workunit and sending it in like other projects?

Thanks!

<IMG SRC="http://boinc.mundayweb.com/one/stats.php?userID=376&amp;prj=4&amp;trans=off"><BR>
<A HREF="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/team_display.php?teamid=1886"><IMG SRC="http://www.fandom.ca/bucket/bcboincrib.png"></A>

I guess the culprit is that the trickles are pretty far apart, so the BOINC "exponential decay" used for RAC, I use an "average trickle time" which in your case tails off to 0 because the first trickle was last October ('04) so the "avg time" is months ago. One of these days I'll come up with a better algorithm for a cpdn workunit for RAC.

Prior to receiving a sulphur model on 9/1, my average (sec/TS) for a trickle was around 2.50. I was uploading three trickles a day.

With the sulphur model, my average (sec/TS) for a tricle is 4.56 ~ 4.60, and it looks like I'm uploading one trickle every other day.

The net result is the longer I process a sulpur model, my Recent Average Credit continues to plummet. Have you or are you making any plans to change the rac algorithm? Barring that, is there a way to dump the sulphur model I'm running and get something more suited to my system?

I don't think folks should be penalized for running this model.

RAC is a pretty meaningless measurement. The closest analog we have been able to come up with is speedometer (RAC) vs. odometer (Total Credit). Because the calculation for RAC is so time dependent, anytime you have a change in the pattern, you will see changes like this.

There is no penalty. It is just the way that RAC is calculated. We beat this to death in the Wiki if you want to read up on it.
<p>
<a href="http://boinc-doc.net/boinc-wiki/index.php"><b>BOINC-Wiki</b></a>
<img src="http://www.boincstats.com/stats/banner.php?cpid=a6477942e70ed39f669d1ff2ede05be8">

82% slower on sulphur is unusually slow. Most people are seeing 60% slower. Parameters can make a difference - normally not more than about 10%. So it seems some computers suffer more than others on changing to sulphur.

I think the project is keen to get some sulphur models back by Jan 06, so I don't understand why there is only 50% extra credit for something that takes 60% longer. There was some talk on the alpha board suggesting a bonus at the end but I haven't seen anything to suggest this has or will be implemented.

I think the project would prefer people to stick with the sulphur model if possible. It is possible to get a slab model by making only limited space available but I would hope most people would want to do whatever helps the project most rather than chasing credits at the expense of what the project wants.
_______________________________
Visit <a href="http://boinc-doc.net/boinc-wiki/index.php?title=Climateprediction_FAQ">BOINC WIKI</a> for help

And join <a href="http://www.boincsynergy.com/">BOINC Synergy</a> for all the news in one place.

we could always bump up the "credit per trickle" in the database if we deem that we're "under-crediting" -- we just sort of thought a 50% increase was what it was, but if it's indeed 60-70% more we can bump the "credit per trickle" up (and since we recalc credits twice a day it will then get in the next update).

&gt; we could always bump up the "credit per trickle" in the database if we deem
&gt; that we're "under-crediting" -- we just sort of thought a 50% increase was
&gt; what it was, but if it's indeed 60-70% more we can bump the "credit per
&gt; trickle" up (and since we recalc credits twice a day it will then get in the
&gt; next update).

It is indeed more than 50%. In Alpha, Thyme and I calculated between 60 and 65%. Looking at the latest Beta, it appears to be pretty much the same as in Alpha.

&gt;82% slower on sulphur is unusually slow. Most people are seeing 60% slower. &gt;Parameters can make a difference - normally not more than about 10%. So it
&gt;seems some computers suffer more than others on changing to sulphur.

I have several P4 HT's running both models at the same time.
The results I am seeing so far are:

P4 3.6 HT 69%
P4 3.0 HT 87%
P4 2.8 HT 83%

My main PC (3.2GHz P4, H/T on) is running HadSM3 v4.13 at ~2.7 s/TS alongside sulphur at ~4.7s/TS, ie sulphur 70-75% slower than slab when running together.

Pete

&gt; My main PC (3.2GHz P4, H/T on) is running HadSM3 v4.13 at ~2.7 s/TS alongside
&gt; sulphur at ~4.7s/TS, ie sulphur 70-75% slower than slab when running
&gt; together.

I have only one sulfur cycle at the moment also, but my numbers are similar:

2.51 s/TS vs. 4.51 s/TS; and
~20 days vs. 97+ days

That is on a P4 3.2 GHz.

I am not sure what this all means. But, for people that run more than one project I have to admit that it is possible that the deadline is also unrealistic. This is the first time I have seen CPDN in EDF for a work unit. I did up my resource share, with the unfortunate side effect that I downloaded another slab model.

Time will tell what this means for CPDN.
<p>
<a href="http://boinc-doc.net/boinc-wiki/index.php"><b>BOINC-Wiki</b></a>
<img src="http://www.boincstats.com/stats/banner.php?cpid=a6477942e70ed39f669d1ff2ede05be8">

Win XP on 3.2 HT P4 Running one of each ... diff. 77%

Win XP on 3.6 HT P4 Currently running 2 sulphur, but compared with completed ... diff. 65%

HTH

&gt; Win XP on 3.2 HT P4 Running one of each ... diff. 77%

Yes, but your slab/hadsm3 model of the pair is running quite a bit faster than they did when you had two slab models running under hyperthreading. Over the last two trickles, it is averaging about 2.84 sec/Ts whereas before they were averaging 3 to 3.3 sec/TS when two slab were running. We saw this in alpha/beta sulphur as well, where when running both slab and sulphur, the slab would be speedier than expected, at the expense of sulphur. So your throughput should be about what one would expect.

For comparison with all those HT Pentiums, my stock Athlon XP3000+ is currently doing 4.83 sec/ts

With the Slab model it ran around 2.98 sec/ts

That is a 62% slow down.

---

4.83 sec/ts

I am not positive what the s/ts means. I am thinking Seconds per Time Slice, but I would like a little better clarification.

---

seconds per time step ...

---

ok based on these numbers I am bumping up credits for sulphur cycle runs to be 70% more than slab instead of 50%. This will show up in the next stats recalc which will be in about a half-hour.

A bit OT but, to expand a bit, for folks new to CPDN: \"What and why is a Time Step?\"

This applies to HadSM3 and Sulfur Cycle Models. (Coupled Model is different but, not to worry, we won\'t see CM in public release for months, at best.)

A Time Step is a 30-minute time-slice of processing for many parameters, in Model time, for the entire atmosphere. Every sixth TS requires more time because there are added heat-flux calculations -- short-wave solar radiation in, long-wave terrestrial radiation out.

Until the CM goes public, the ocean is simplified. Currents flow but there is no turn-over (vertical motion) in ocean calculations. (That goes a long way to explaining \"Cold-Equator\" Runs.) Consequently, TS are atmospheric repetitions through time for the parameter set for the particular Model and Phase.

Each TS processes 7,008 grid cells per level, for each of 19 atmospheric levels. (73 Latitude by 96 Longitude) Of course, the cells are not treated in isolation. Calculations must consider what occurs in surrounding cells, in both the horizontal and the vertical.

There are 259248 TS per Phase (15 Model years), organized into 24 groups of 10,802 TS per Phase. (Three Phases for HadSM3, five Phases for Sulfur Cycle.) After each 10,802 TS, a \"Trickle\" is sent to the server. Twenty-three Trickles do little more than tell the servers that the Model is alive and well; the 24th Trickle contains summary information for your Model -- used to generate the graphs you see on the Project Stats page and in Your Account.

A lot of work is done for each Time Step.

No wonder it takes so long, eh?

(This is surely documented elsewhere but I was too lazy to search for it.)

---

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.

ok based on these numbers I am bumping up credits for sulphur cycle runs to be 70% more than slab instead of 50%. This will show up in the next stats recalc which will be in about a half-hour.

Thanks Carl thats great news

Thank you Carl, much appreciated. Was surprised to see extra credit when I looked today.
On my P4 2.4 the times are
1st Slab unit ~2.72 Seconeds/TS
2nd Slab unit ~2.94 Seconeds/TS

1st Sulpher unit ~4.64 Seconeds/TS

From those numbers 70% more for the Sulpher is almost exact from my first Slab unit.

Any idea on how the times will be on the new units to come out in Feb. 06? 5 Phases like the Sulpher?

Ray Brown

---

Keep on crunching Pizza@Home

Ray
I suggest a search, perhaps on the community forum.
I\'m fairly sure that Carl said somewhere that the coupled model will be one long 50 year run.
No phases, but large trickles all through the model.
More detail around New Years, I\'d guess.
It\'s really going to sort out the stable computers from the rest.