[Knightrider]

Hi,

Thought I would start this thread so as to get all CPDN stuff in one place. So anything CPDN can be posted here.

Trickles: Currently having a problem with the CPDN server not updateing trickles. From what I can gather this is an ongoing problem, so I will keep crunching my current wu's until it is sorted out one way or another. This thread dicusses the problem.

http://www.climateprediction.net/boa...forum.php?f=12

Backup: This is something that had not occured to me and given the long runs I think this is an excellent idea. I was prompted by this thread:

http://www.climateprediction.net/boa...pic.php?t=2377

K.

[Knightrider]

Backup: So I had to try the backup idea and.......you guessed it.....didn't work.

What happend is that my ichat failed to open. All it does is bounce the opening window on the screen once, it will not stay open and drops out. So I tried various things ending up in having to do a reinstall, twice.

I backed up CPDN on to two DVD's, it being so big a folder. When I restored it it would not work and I still had the ichat problem so I abandoned it straight away and did my second Tiger install.

I reload BOINC Menu bar and downloaded fresh wu's. Two would not come down and were lost, but now have four running.

In the backups I had reached phase two. One thing I had noticed was that after phase two started, the trickles, which were still being sent, were not showing up as credits on my wu account pages.

ichat still not working.

Hey Ho, the wind and the rain.

K.

[Knightrider]

From the news page:-

"Due to routine maintenance work in Computer Services, the climateprediction.net servers will be down for some or all of 7-9 January. This may affect communication with the project in that period. We're sorry for any inconvenience."


K.

[Snake_doctor]

Originally Posted by Knightrider

From the news page:-

"Due to routine maintenance work in Computer Services, the climateprediction.net servers will be down for some or all of 7-9 January. This may affect communication with the project in that period. We're sorry for any inconvenience."


K.

The question of course is CPDN smart enough to store and upload ALL the trickles after such an outage? Humm... Guess we will know on the 10th.

These models are so delicate they fail at the slightest whim.

Regards
Phil

[Knightrider]

Originally Posted by Snake_doctor

These models are so delicate they fail at the slightest whim.

They are! After phase 1 I could not upload 3 out of 4 of the results zip files.

The error msg:

[climateprediction.net] Error on file upload: [Sulpher model xxxxx] locked by file_upload_handler PID=5848
[---] Network error: a timeout was reached

For each of the three. I tried every which way to sort it out, but to no avail. All three continued to be crunched with the upload attempt repeating every 2 or 3 hours for each wu. Last night the server went down and now my computer pages have gone Hope I didn't crash the server


So I have given up temporarily, and switched to Einstien and the new 'Albert' wu's. If a new version of CPDN for Macs comes out I will go back.

K

[Snake_doctor]

Originally Posted by Knightrider

They are! After phase 1 I could not upload 3 out of 4 of the results zip files.

The error msg:

[climateprediction.net] Error on file upload: [Sulpher model xxxxx] locked by file_upload_handler PID=5848
[---] Network error: a timeout was reached

For each of the three. I tried every which way to sort it out, but to no avail. All three continued to be crunched with the upload attempt repeating every 2 or 3 hours for each wu. Last night the server went down and now my computer pages have gone Hope I didn't crash the server


So I have given up temporarily, and switched to Einstien and the new 'Albert' wu's. If a new version of CPDN for Macs comes out I will go back.

K

I have had about 12 models fail at various points during processing. 5 of those failed to even start when I first attached to the project. I had one fail at about 60% (really yanked my chain). But I am determined to finish at least one model.

My current one is at 40% complete and I would hate to think that the project is going to do something that will screw it up. I have only had 2 sulphers and they both blew off after about 20 %. The Slabs are ok for now for my normal systems, but I have a G5 Dual that really sings with the sulphers, but I would rather us it on medical research until the CPDN folks can build some reliability into the system.

Regards
Phil

[Knightrider]

Here is the text of a message posted on Einstien by Bruce Allen, which seems to explain the file upload problem:-

"The file uploading problems are due to an emergency change that David Anderson made in the BOINC client a couple of months ago when SETI BOINC was having trouble with its servers. Unknown to me, he modified the client so that if the upload file size is less than one megabyte, the client always tries to upload the file starting from the begining, rather than doing partial uploads.

If one of these uploads failed, then the next attempt would cause problems on the server side.

After some discussion, David has reverted this change: future clients won't suffer from this problem.

As a temporary workaround for people using current clients, I have modified the Einstein@Home server-side upload code to eliminate the failure mode.

People with slow connections who have trouble completing uploads should upgrade their BOINC client as soon as a new release comes out. 5.2.13 is flawed in its handling of file uploads.

Bruce"

The latest version on BOINC Dev is 5.2.15 so I will return to CPDN useing that, when my current Einstien work units dry out.

K.

[Snake_doctor]

Originally Posted by Knightrider

Here is the text of a message posted on Einstien by Bruce Allen, which seems to explain the file upload problem:-

"The file uploading problems are due to an emergency change that David Anderson made in the BOINC client a couple of months ago when SETI BOINC was having trouble with its servers. Unknown to me, he modified the client so that if the upload file size is less than one megabyte, the client always tries to upload the file starting from the begining, rather than doing partial uploads.

If one of these uploads failed, then the next attempt would cause problems on the server side.

After some discussion, David has reverted this change: future clients won't suffer from this problem.

As a temporary workaround for people using current clients, I have modified the Einstein@Home server-side upload code to eliminate the failure mode.

People with slow connections who have trouble completing uploads should upgrade their BOINC client as soon as a new release comes out. 5.2.13 is flawed in its handling of file uploads.

Bruce"

The latest version on BOINC Dev is 5.2.15 so I will return to CPDN useing that, when my current Einstien work units dry out.

K.

Some one need to tell our optimization guys about this. The current MacNN optimized client is based on 5.2.13.

Regards
Phil

[Knightrider]

Re Upload problem.

Ok - continue to have a problem uploading to CPDN, uploads to Einstein go ok.

I have come to the conclusion that either BOINC or the server, or both, cannot handle uploading multiple results within a short time frame +/- 10 minuets at least.

Bearing in mind that the CPDN zip upload at the end of phase one is about 8.15 mb. The first of the four on my quad went up ok. The following two, about 24 hrs later, wanted to go up at the same time but both failed. The fourth one went up ok about 24/36 hours later. I had spaced the wu's to about 24hrs between each one, but because you cannot know when a phase will complete, two results came up together. I have now increased the spacing to 36 hours between the first two wu's and a third is about 140 hours behind them.

I had noticed that when I first downloaded four wu's and seperated them all out to about four hour gaps, they eventually all adjusted to run at the same % rate and cpu time, consequently, with a good chance of 2 or 3 of them finishing phase one at the same time, which is what happen'd first off.

Therefore I conclude that one wu should be able to run without trouble and I recommend spacing out 2, 3, and 4 to at least 36 hours between each wu.

Trickles are not affected and seemed to go up ok throughout.

K.

[Knightrider]

Of all the distributed computing projects available at the moment; I think CPDN is one of the most important, if not the most important.

Here's why:-

http://news.bbc.co.uk/1/hi/sci/tech/4660938.stm

K.