climateprediction.net home page
New App is much slower/less efficient

New App is much slower/less efficient

Questions and Answers : Macintosh : New App is much slower/less efficient
Message board moderation

To post messages, you must log in.

AuthorMessage
Billy

Send message
Joined: 23 Jan 07
Posts: 26
Credit: 852,233
RAC: 0
Message 35625 - Posted: 2 Dec 2008, 14:59:10 UTC

I was running a HADCM3 app 5.39 and getting 1.9 sec/TS. Now the newer HADCM3 app 6.04 is running at 2.9 sec/TS.

This seems to be a big step backwards for the newer app.

Cheers.
ID: 35625 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 35627 - Posted: 2 Dec 2008, 19:44:24 UTC

The longer time is due to altered code to make the models more stable.
And different starting parameters can also affect the overall speed of different models.


Backups: Here
ID: 35627 · Report as offensive     Reply Quote
Billy

Send message
Joined: 23 Jan 07
Posts: 26
Credit: 852,233
RAC: 0
Message 35628 - Posted: 2 Dec 2008, 20:13:59 UTC - in response to Message 35627.  

The longer time is due to altered code to make the models more stable.
And different starting parameters can also affect the overall speed of different models.

I am glad the new code is improved to give more stabillity. Is there more credit given to adapt to this change in the code?
ID: 35628 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 35629 - Posted: 2 Dec 2008, 21:12:11 UTC

Yes. When the HADCM was optimised the credit per trickle was increased.

For most projects \'optimised\' means faster, but for CPDN the most important thing is model stability. That\'s because the models are so long and perform so many complex calculations.

Your new models are also HADCMs but they\'re for different research - the geoengineering experiment.
Cpdn news
ID: 35629 · Report as offensive     Reply Quote
Billy

Send message
Joined: 23 Jan 07
Posts: 26
Credit: 852,233
RAC: 0
Message 35638 - Posted: 3 Dec 2008, 16:47:35 UTC - in response to Message 35629.  

Yes. When the HADCM was optimised the credit per trickle was increased.

For most projects \'optimised\' means faster, but for CPDN the most important thing is model stability. That\'s because the models are so long and perform so many complex calculations.

Your new models are also HADCMs but they\'re for different research - the geoengineering experiment.
According to the first few trickles I have sent in, the credit per trickle has not increased, but is still 311 credits per trickle. For my computer, this means a 55% decrease in credits with the optimised application. Is something wrong?

Cheers.
ID: 35638 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 35640 - Posted: 3 Dec 2008, 18:05:27 UTC
Last modified: 3 Dec 2008, 18:19:43 UTC

Hi again Billy

Les must have thought you\'d been running your previous model for much longer than was in fact the case. I\'ve looked at its web page and see that you downloaded it in August 08, which is after the HADCMs were optimised for all platforms. The applications page shows the current 80-year HADCMs launched for the Geoengineering experiment on 22 Oct, but the previous HADCMs were also the optimised version. I know for sure that this was the case for both Windows and Linux, and from what you are saying it must have been the case for Macs as well. This is why your credits haven\'t increased per trickle.

I can\'t explain why your current model is so much slower than the one that crashed except that individual models do vary, occasionally quite a lot. I haven\'t got a Mac, but I\'ve looked at the speed of my current Windows Geoengineering model and have compared it with how fast an earlier optimised HADCM ran on the same computer. The speeds are similar. But the same computer has crunched HADSMs at very different speeds.

Your current HADCM is definitely about 30% slower after 3 trickles than your previous model.

If the credits are important to you, here are some suggestions:

* You seem to have downloaded two HADCMs but you\'re only running one. What about suspending the model you\'re running and letting the other run to see how it performs?

* If your other new HADCM runs no faster, in the project preferences section of your account, select HADSM or HADSMMH or both. Deselect HADCM. Abort your current HADCMs and see whether the HADSMs crunch at the speed you like.

* Or, if you\'re reluctant to abort your new 3-trickle model(s) before knowing how the HADSMs will perform on your Mac, make a backup of the complete BOINC folder (or BOINC Data folder if you have BOINC v.6), abort the two new models and try the HADSMs. If you don\'t like how the HADSMs perform you could abort them after a couple of trickles (give them time to settle into a rhythm) and restore your backup with the HADCMs. The server would still allow you to continue one or both. If you want to do this we can point you to easy backup and restore methods.

I\'m sorry, I can\'t think of any other solutions.
Cpdn news
ID: 35640 · Report as offensive     Reply Quote
Billy

Send message
Joined: 23 Jan 07
Posts: 26
Credit: 852,233
RAC: 0
Message 35642 - Posted: 3 Dec 2008, 20:05:00 UTC - in response to Message 35640.  

Hi again Billy

Les must have thought you\'d been running your previous model for much longer than was in fact the case. I\'ve looked at its web page and see that you downloaded it in August 08, which is after the HADCMs were optimised for all platforms. The applications page shows the current 80-year HADCMs launched for the Geoengineering experiment on 22 Oct, but the previous HADCMs were also the optimised version. I know for sure that this was the case for both Windows and Linux, and from what you are saying it must have been the case for Macs as well. This is why your credits haven\'t increased per trickle.

I can\'t explain why your current model is so much slower than the one that crashed except that individual models do vary, occasionally quite a lot. I haven\'t got a Mac, but I\'ve looked at the speed of my current Windows Geoengineering model and have compared it with how fast an earlier optimised HADCM ran on the same computer. The speeds are similar. But the same computer has crunched HADSMs at very different speeds.

Your current HADCM is definitely about 30% slower after 3 trickles than your previous model.

If the credits are important to you, here are some suggestions:

* You seem to have downloaded two HADCMs but you\'re only running one. What about suspending the model you\'re running and letting the other run to see how it performs?

* If your other new HADCM runs no faster, in the project preferences section of your account, select HADSM or HADSMMH or both. Deselect HADCM. Abort your current HADCMs and see whether the HADSMs crunch at the speed you like.

* Or, if you\'re reluctant to abort your new 3-trickle model(s) before knowing how the HADSMs will perform on your Mac, make a backup of the complete BOINC folder (or BOINC Data folder if you have BOINC v.6), abort the two new models and try the HADSMs. If you don\'t like how the HADSMs perform you could abort them after a couple of trickles (give them time to settle into a rhythm) and restore your backup with the HADCMs. The server would still allow you to continue one or both. If you want to do this we can point you to easy backup and restore methods.

I\'m sorry, I can\'t think of any other solutions.
I was referring to the application change from 5.39 to 6.04 as I was getting similar performance with HADCM model downloaded in April 2007 as the one from August 2008. Perhaps I should be thankful as I have read in another thread that newer Macs are crashing immediately trying to run the new HADCM models. I will however take your advice and investigate how the other models perform on my computer. I am running several projects and thus only one CPDN model at present.
ID: 35642 · Report as offensive     Reply Quote

Questions and Answers : Macintosh : New App is much slower/less efficient

©2024 climateprediction.net