By climateprediction.net I got task demanding 13 hours of computing a day to make it on time.
By SETI@home I just got some new type of SETI "Astropulse" task demanding almost 7 hours of computing a day to catch the deadline.
All that with nonstop 90% CPU usage, 90% RAM usage at 100% CPU frequency.
I know, that if I do not like it I need not to do it, because there is probably superfluous computing capacity in these projects, but wasn't this once supposed to be fun?

I agree. Some of the projects have quite heavy duty workunits with very short deadlines. And yes, this is not funny sometimes....

On the other side, some projects need to send out heavy duty workunits because only if a complete simulation can be run through distributed computing makes sense - for example, World community Grid's "Computing For Clean Water" (AFAIK).

Same is probably the case for climateprediction.net. Well, 13 hours is really good for CP, I remember getting workunits which would have taken 14 days (!) to complete on my Athlon 2600+ a few years ago..... of course I aborted them. And it drove me away from climateprediction.net, too, then.....

But I think you found the wrong place to complain; you should write to the project's managers and ask for a longer deathline. In fact, only there you can (and hopefully will) be helped - even with absolutely optimized code, BOINC will need its time to do its work, and only the projects can give you the time.....

Thank you for sympathy. I didn't mean to whimper. I'm just a little saddened, that the original idea of spare time crunching was effectively put away.

My BOINC is only allowed to do work between 9pm and 7am, when the electricity rates are lower, the rest of the day it's suspended at 'time-of-day'. It gets work from 7 projects, has multiple tasks per project in cache (bar T4T and Surveill) --reasonably, connect to of 0.50 days and additional cache of 1.00 days. The projects in question are for this month: World Community grid (Clean Water), Spinhenge, Einstein, Rosetta, Constellation, Test4Theory and Surveill.

The system itself is an i3-530, 2 core CPU with hyper threading, 4Gb RAM, Windows 7 64bit. It's allowed to use all cores. It manages to do all the work that is cached well within the time allotted.

I have another P4 3.0GHz which acts as my TV server. It's only on when we are watching something we downloaded, lately the Babylon 5 series. It's also only allowed to do work between 9pm and 7am, while it's attached to 2 projects, Climateprediction.net and Primegrid. Although it's a system with hyper-threading, I have set BOINC to run only on 1 CPU (use at max 50% of processors).

We shut this system down when we stop watching whatever it is we watched, so it differs every evening how long the tasks can run, there's days it ran 2 hours, days it ran 4 hours and in between or over. The system isn't on 24/7. It even happens that this system stays off for more than a day.

Thus far I haven't seen any problems on either system.
Now, luckily, CPDN does not adhere to their deadlines. At least not for the models that have a deadline of a year or more. Even while BOINC may be complaining that such a task is past its deadline and that you better abort it, you can ignore that message and continue running it to completion and still get credit.

Seti's Astropulse tasks may tell you they take that long, but in reality they don't. The initial time to completion on all tasks is an estimated time. During the running of the task, BOINC will learn how long they actually take. While Astropulse tasks may start out at 480 hours estimated run time, they run in something like 140 hours on modern machines. Quicker even when you use Seti's optimized applications, or the GPU applications (see their forums for more information on that, or ask there).

---

Thanks for encouraging answer.
I will post the real computing times and a percentage of "real computing time"/"assigned time", when workunits are done.

Thank you for sympathy. I didn't mean to whimper. I'm just a little saddened, that the original idea of spare time crunching was effectively put away.

You are welcome. And I didn't take it as whimpering - in fact I think it is a point well worth debating over. And moreover, I totally agree that restrictive deadlines are against the original idea.


@ Ageless: Interesting - both your systems and the info about CPDN and Astropulse. I know that a few other projects don't adhere strictly to their deadlines too, so even if you get over it they will accept the results. But nevertheless, deadlines have of course their eligibility.


Have a nice Sunday, both of you :)

The CPDN models that are really long are called Hadcm3n. Members can go into their CPDN account and then into their CPDN preferences. They can edit the type of models they want and don't want. After editing the preferences one should open Boinc manager and update the CPDN project in the projects tab; this tells the server about the changed preferences.

The deadline for these Hadcm3n models is rather short for a reason. They are 40 climate years long (that's simulated years, not real-world years...). You see _40_ in the model name. But each is part of a 160-year simulation. When the first 40-year simulation of each series sends its data back to the server, the next 40-year period can use that as its starting data.

The alternative would be for each computer to process the whole 160 years. 40 years may not be universally popular, but I can assure you that 160 years is not universally popular either.

Researchers in the Dept of Oceanography at the University of Southampton need results from the completed 160-year series as soon as possible. They have an array of sensors suspended in the Atlantic from N America to Africa to try to find out whether the ocean current is slowing down. But they also need the climate model simulations. They can't complete their research until they get data for the completed 160-year periods back.

Crunchers who want shorter models can instead select the Hadam3P regional types. Of the 3 regional types the shortest of all is Southern Africa. Unfortunately, at the moment there aren't any of these available for download but there may be soon.

Thank you all for the interest and useful information.
I didn't know about the option in the CPDN.
Astropulse on the first machine really did come to sense (estimated cca 80 hours instead of originally predicted 400 something).

My machine No.2:
ASUS Notebook X58L Series
MS Vista Home Premium 32-bit SP2
Intel Pentium Dual CPU T3200 @ 2.00GHz, 2,0GB RAM,
GPU: Mobile Intel(R) GMA X3100 (448 MB)

All calculations for (90% CPU time, 90% swap space, 90% memory, 100% CPU frequency)

Workunits (4 228 962 GFLOPs each)
1) hadcm3n_o7fv_1940_40_007444247_2
2) hadcm3n_o6dw_1940_40_007444039_0

Granted times (thanks to http://www.timeanddate.com/date/duration.html)
1) From 9 Sept 2011 8:27:00 Till 9 Dec 2011 14:56:43 ie. 7 885 783 seconds
2) From 9 Sept 2011 9:27:42 Till 9 Dec 2011 15:57:25 ie. 7 885 783 seconds

After some 30hours of crunching it looks like this:

Workunit 1) 2.891% in 29:54:40
ie. 122259.29142 GFLOPs in 107680sec giving CompPower of: 1.114 GFLOPs/sec

Workunit 2) 2.77% in 29:18:27
ie. 117522.85398 GFLOPs in 105504sec giving CompPower of: 1.135 GFLOPs/sec

Necessary computing time in this configuration:
1) 4 228 962 GFLOPs/1.114 GFLOPs/sec = 3796195,7 seconds (which is 48% of Granted time ie. cca 11.5 hours of computing each day to meet the deadline)
2) 4 228 962 GFLOPs/1.135 GFLOPs/sec = 3725957,7 seconds (which is 47% of Granted time ie. cca 11h and 17min of computing each day to meet the deadline)

OK, so I've learned my lesson. Night shifts for present tasks and change the model preferences for the next.

Thank you again for help.

After 46 hours of computing:

1st Task: 45h50m14s/4.772% = 960h32h43s total computing time, which is 43.8% of granted time.

2nd Task: 45h12m42s/4.645% = 973h20m26s total comp. time making 44.4% of granted time.

Somehow, it´s getting better. Or my math is sloppy...?