I getting a wee bit peeved with CPDN Boinc. I\'ve tried it numerous times with numerous problems. I\'ve only had one completed unit. I\'ve had units die after 80% done, after 10% done, whenever. I just got a new, faster laptop and decided to give it a go, 2.1GHz Centrino with 2Gb RAM. Here I am 7days/ 36hrs on CPDN/ 6.5% done, no trickles, not one. Does anyone know what\'s going on, should I just give up on CPDN.

I like the idea of CPDN, it\'s great to have around when the other Boinc projects are having problems, really credits mean nothing, but they are nice to have around....

Slainte, in exasperation,

David

server problems are being sorted
Tue, 21 Jun 2005

We would like to apologise for the server problems that we have had over the last few days. The immediate problem was due to a server which had filled up, this is currently being sorted out. With both Tolu and Neil back from holiday, we hope to have the remaining issues sorted out as soon as possible. We also hope to have another software engineer working with us shortly, so this very unusual lack of computing support should not happen again. With thanks to all our participants for their patience.



source: http://www.climateprediction.net/newsb.php

I hope that this helps explain why you're having a problem. I've got 2 wu's for CPDN and I understand. Hopefully they will have it fixed soon.
<img border="0" src="http://boinc.mundayweb.com/one/stats.php?userID=1343" />

Hi, david.
The trickle server is down, but according to the dev team, it should work again tomorrow.
see <a href="http://www.climateprediction.net/board/viewtopic.php?p=27808#27808">here </a> and thanks for your patience...(don't give up just for credit problems :o))
-----------------------------------------------
<a href="http://www.boincforum.info/boinc/">boincforum</a>

OK, since your nice guys, I'll stop moaning, afterall compared to SETI, LHC, etc, the CPDN servers to tend to be pretty stable.

One other question. During one of my many failed work units I did see a rather pretty CPDN logo on the screensaver, it not there anymore, what happened to it?

Maybee it was replaced in a newer version of hadsm. I don't know whether it really was and when in what version it was done.

davidmcw said:

&gt; OK, since your nice guys, I'll stop moaning, afterall compared to SETI, LHC,
&gt; etc, the CPDN servers to tend to be pretty stable.
&gt;

I noticed that also with the other projects while looking at LHC (no work being given out ATM as the project is still in Beta phase) the Einstein boards (the same kind of problems) for some split project work so have decided to stick here.

I had just began to wonder with approaching 100k runs of the simple HadSM3 model and non of the later stage experiments ready for general release yet, whether each run was still giving as much useful information. Just for comparison, some of the recent problems, though frustrating (mainly through the lack of feedback) are nothing compared to those which used to happen when the only other distributive computing project I have ever run (United Devices) first came on line 4 years ago. Things were so bad with final uploads at one time, I had a batch file running just to back up the running model automatically every 2 hours or so so that if an upload failed, I'd got various backups I could try to upload manually.

Pete

100k runs is less than 8 simulation for each model version. Add in 30k trickling machines and we are working towards an initial conditions ensemble size of 10.

I have been expecting to see models with initial condition parameter of greater than 0.09. I don't know what difficulties or priority is given to sorting out sensible start field and getting this coded up ready for distribution.

The impression I have though is that they want ic ensembles of more than 10. My reasoning for this is:

If they were not particularly interested in ic ensembles of more than 10 and the start fields were not readily available, it would be fairly easy to add a 4th value for some of the parameters where the response appears non-linear. I am sure this would be of some use. Therefore I think it is likely that they really do want at least 10 simulations of each model version.

There has to be a point were the marginal value of increasing the sample size just slightly suffers from diminishing returns, but I don't think we have reached it yet. They were talking about an initial condition ensemble size of several hundred for the standard model. I could also imagine wanting a sample of a few hundred model versions and wanting ic ensembles of at least 30 for each of those model versions.

HTH

Also, 100k is the total. A percentage of these may well be rejected from the final group of those accepted. There was a similar 'weeding out' for those used for the paper earlier this year.
So, perhaps double this number?
After all, a lot of machines are having trouble running this experiment and may find the next one even harder to complete, but their owners will want to keep running on this project. So keeping this one going will ensure that an even better selection will gradually become available, even after experiment 2 starts.