I seem to be getting much longer gaps between trickles with sulphur wu than with slab - like a factor of around 2x - 3x as long. Same applies to sec/timestep.

Are other people seeing this?

My impression too is that the effect is more pronounced on linux than on windows - anyone else noticed this?

I have not produced exact figures yet, just a feeling from the sort of numbers I have been getting.

Anyone who has completed a sulphur wu, do they have 3 stages each of 25 trickles, like the slab models? Each trickle seems to be ~10800 timesteps, just like with the slab model.

If I am right that the model is much slower no doubt it is doing even more - I am not complaining here! I would however like to suggest that it would be good to restrict sulphur models to fast machines which are on a lot. I believe the BOINC infrastructure allows WU to be restricted to certain cpu speeds, not sure if you can manage to restrict by on-time.

I have two sets of machines, one set is on 24/7 but have ~700 MHz cpu. The other - owned by a local charity and used by permission, have 2.8 MHz cpu but are only on around 37 hours/week. In terms of throughput per week they are roughly similar - taking about 15 wk for a slab model, which leaves plenty of slack for the 49 week deadline. A model taking 3x as long would be taking 45 out of the 49 weeks - uncomfortably near the limit.

If you can\'t segregate out the sulphur models for the fast always-on machines I will need to move a lot of my BOINC-ing to other projects.

---

Yes, Sulphur cycles WUs takes longer - both in term of sec/TS and phases (3 vs. 50). A factor of about 2.5x is correct. Credit reward respect that as well: 6805.26 for slab vs 19281.57 for sulphur

Some basics about Sulphur Cycle Experiment
http://www.climateprediction.net/science/s-cycle.php

Much more on phpBB forum...

---

<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a>

Thanks Honza.

Yes, Sulphur cycles WUs takes longer - both in term of sec/TS and phases (3 vs. 50).

I think you mean 5 phases? But how many trickles in each please? Or how many in total for the whole run? I am trying to guesstimate the time for the whole run from the rate of trickles.

A factor of about 2.5x is correct. Credit reward respect that as well: 6805.26 for slab vs 19281.57 for sulphur

which is fine unless the credit goes to zero after the deadline is reached...

---

I don\'t understand, 50 phases?

Was a typo. Should be 5 phases.

How many trickles altogether please? I am trying to work out if there is any prospect of these wu finishing on my relatively slow boxes, and on the fast boxes that are only on in working hours.

In sulphur, 5 phases x 24 trickles/phase = 120 trickles

which is fine unless the credit goes to zero because the deadline overruns...

It does not go to zero past deadline.

In sulphur, 5 phases x 24 trickles/phase = 120 trickles

thanks geophi

---

Hi Gravywavy,

It looks that Honza made a typo, but:
every phase has 24 trickles
-Slab has 72 trickles in 3 phases
-Sulphur has 120 trickles in 5 phases.

Sulpher does extra calculations and has more model-years, so takes considerably longer.

In case you are running Sulphur Cycle version 4.22, you will posibly be affected by performance degradation, specially if you are running slow cpu, intel and linux.

Please see:

http://climateapps2.oucs.ox.ac.uk/cpdnboinc/forum_thread.php?id=3552

or

http://www.climateprediction.net/board/ forum

Btw, we have met before on e@h, how is that internet cafe project keeping?

---

Btw, we have met before on e@h, how is that internet cafe project keeping?

yes, it\'s me again!

The internet cafe is the same as the local charity I mentioned in this thread - the cafe is aimed at homeless, drug users, people with mental health issues, etc. Cheap food, free net access and teaching available to gain/improve IT skills.

We got off to a good start, switched from e@h to climate on the cafe after they upgraded all their machines and the slab models looked runnable (est 10-15 wks to complete on a 2.8GHz box that is only on 32-37hrs/week).

The sulphur model does not look runnable in that situation - it would be getting on for a year to run it. The potential for a change in policy or a change in the manager could mean that BOINC is removed overnight. OK for a wu lasting three months but not one lasting ten months, in my opinion. The chance of it happening increases with time and so does the amount of work at stake (the project would lose any scientific advantage from the work units).

That is why I am hoping the project can figure out a way to allow us to opt out of sulphur and go on getting plain old slab models, leaving the exciting stuff to those with fast boxes on 24/7.

---

Was a typo. Should be 5 phases.

Yes, was a type - sorry for that, thanks for correction [too late to edit].

---

<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a>

Btw, we have met before on e@h, how is that internet cafe project keeping?

yes, it\'s me again!

Probably quite belated, but Welcome to Climate!

The internet cafe is the same as the local charity I mentioned in this thread - the cafe is aimed at homeless, drug users, people with mental health issues, etc. Cheap food, free net access and teaching available to gain/improve IT skills.

A Noble goal, not all can keep up with the system and will be overwalked by it, hopefully will gaming or IT skills bring some back on the rails or at least alliviate the pain.

We got off to a good start, switched from e@h to climate on the cafe after they upgraded all their machines and the slab models looked runnable (est 10-15 wks to complete on a 2.8GHz box that is only on 32-37hrs/week).

The sulphur model does not look runnable in that situation - it would be getting on for a year to run it. The potential for a change in policy or a change in the manager could mean that BOINC is removed overnight. OK for a wu lasting three months but not one lasting ten months, in my opinion. The chance of it happening increases with time and so does the amount of work at stake (the project would lose any scientific advantage from the work units).

That is why I am hoping the project can figure out a way to allow us to opt out of sulphur and go on getting plain old slab models, leaving the exciting stuff to those with fast boxes on 24/7.

Some forum members have questioned the friendlyness of the download module. The choise slab vs sulphur is one of the issues in that.
As outsider I don\'t know the status of the slabs, e.g. are the still usefull or where we rerunning the same paramsets on different platforms, but if still usefull, users should have the choise to run slabs.
Continuous pressure from forum members might change the present policy.

---

Some forum members have questioned the friendlyness of the download module. The choise slab vs sulphur is one of the issues in that.
... users should have the choise to run slabs.

actually we do have the choice, if we go back to classic cpdn we only get slabs

Continuous pressure from forum members might change the present policy.

if enough people make the retrograde step, and explain why, then those who are encouraging us to move from classic to BOINC might get the hint...

---

The scientists have enough slab models.
Now they want more advanced data.
What\'s so hard to understand?

The scientists have enough slab models.
Now they want more advanced data.
What\'s so hard to understand?

I explain Carl\'s decision to turn off slabs for the moment was based on the need for a quick batch of 500 or more Sulpher runs.

Please take into account that those slow machines running Sulpher now, will not generate one of the 500 first completed runs.

If it is still usefull, those machines could as well run slabs.

---

Carl posted another message, to the effect that they had now generated a LARGE number of sulphur sets. Can\'t remember if it was 5,000, 50,000, or what.
But it was because the scientists wanted more sulphur data.
THIS was when he said that slab had been turned off, not when the 500 runs was announced.
The new runs have the usual 1 year \'deadline\'.

Carl posted another message, to the effect that they had now generated a LARGE number of sulphur sets. Can\'t remember if it was 5,000, 50,000, or what.
But it was because the scientists wanted more sulphur data.
THIS was when he said that slab had been turned off, not when the 500 runs was announced.
The new runs have the usual 1 year \'deadline\'.

Yes, Les you are right :!: it was a 50k batch (hopefully that batch is not affected by double CO2 absence)

On a P4 2.4G I would prever to run a Sulphur, but before the switch off it would only give me Slabs.

On a PIII 800MHz I got a Sulphur, which supprised me.
Now, it is winter here, I have 8 machines running and keeping my place on temperature. Summer coming I will get a problem, I have to shutdown machines or I will suffocate, besides that I will lose the synergy I have now.

In this situation it makes a differance if I deal with completion times of 1700 hrs or 5200 hrs.
If there is no use for slabs anymore, I have to accept this fact, but if there is still use for slab, than I would prever to have the choise.

I wonder what the resistance is here, boinc client code is not holy or is it. Besides that, with my remarks I only try to improve the user experiance :!:

---

Earlier on, Carl and Tolu tried to limit what could be downloaded by various PCs, trying to make it make sense based on computers downloading the model. However, it apparently did not work. I\'m not sure how one would go about having the software automatically figure out what to hand out based on computer speed and time spent on a project (which in itself can be very misleading).

But if there could be a setting, on the individual computer page perhaps, which would have a dropdown list of model types, like the current dropdown about which preference to use (home, work, etc.), that would be good.

But if there could be a setting, on the individual computer page perhaps, which would have a dropdown list of model types, like the current dropdown about which preference to use (home, work, etc.), that would be good.

I suggested this extension back in BOINC Beta to Dr. Anderson so they could use it for the AstroPulse application. For example, I would go back to giveing the SETI@Home project a much larger percentage of time if I could, if nothing else, \"suggest\" that I would prefer to download and process AstroPulse work units.

This would not have been a mandatory setting for the scheduler necessasarily, but, if there were a choice of work queued up it would let me have some AstroPulse work instead of having a random draw.

For whatever reason he was not supportive of the idea, but, I think it is a natural. Predictor@Home has had some issues with some work unit types and it would have been nice to be able to say \"no thanks\". :)

Maybe we can prevail on Carl to add the feature and push it back into the baseline.

---

One way the project can select which computers get work is by a flag that will cause work not to be sent to hosts that can not complete it before the deadline. As far as I know this is the only current limit on what kind of work is downloaded. I also believe this flag is not set for this project since in most cases (at least prior to sulphur) the project is not too concerned about getting the model back by the deadline.

I would also like to see a project preference for what type of work a client gets. CPDN needs this I think. I also think SETI will need this when astropulse is released. I would be a good option for Africa@home as well (they are currently planning on having seperate projects of which malaria control is the first). WCG and FaH may also benefit from allowing participants to select which sub-project(s) are desired.

I would set it up with a \'use exclusively\' yes/no for the project and an \'allow (app name)\' yes/no for each app. If the exclusive option is set to yes then only if work for the allowed apps is available will work be assigned. If the exclusive option is set to no then the server would look at the allowed apps first and use a different app if no work was available for any of the allowed apps.

---

BOINC WIKI

BOINCing since 2002/12/8

The problem with all this, is that the current and future phases of cpdn are all going to be large.
And people want to have a short, easy to run option, for use on older, slower computers.
So the ability to choose which type of model to run won\'t help these people.

Earlier on, Carl and Tolu tried to limit what could be downloaded by various PCs, trying to make it make sense based on computers downloading the model. However, it apparently did not work. I\'m not sure how one would go about having the software automatically figure out what to hand out based on computer speed and time spent on a project (which in itself can be very misleading).

I am very surprised.

Other projects servers occaisionally generate the message \"There was work but it won\'t finish in time\". This code takes into accout the % of time the machine is on and the % resources devoted to the particular project.

At present this code does not prevent the download of a single wu to an empty machine, because it is designed to interact with the EDF processing. Where projects have roughly similar deadlines, say up to a factor of five from shortest to longest, this works well.

It does not work in a human-friendly way if one project has deadlins that are 20x to 30x others.

Looking in from the outside it does not seem a big job to change the code specifically for BOINC.

---

I would also like to see a project preference for what type of work a client gets. CPDN needs this I think....
I would set it up with a \'use exclusively\' yes/no for the project and an \'allow (app name)\' yes/no for each app....

This would be nice.

However, it is more than *I* am asking for. What would be sufficient for me is to be able to set a limit on the number of hours work in a single WU. I don\'t have a preference for the sort of science - I trust that all the science CPDN is doing is deeply worthwhile. My only issue is whether the WU can fit into the other needs of the boxes I have access to. So with my machine\'s benchmarks a cut0off of 2000hrs would mean I\'d get slab or similar short runs, but would not get sulphur or similar long runs.

---