climateprediction.net home page
Iceworlds and Slowdown in slab (hadsm3/hadsm3mh) models
Iceworlds and Slowdown in slab (hadsm3/hadsm3mh) models
log in

Advanced search

Message boards : Number crunching : Iceworlds and Slowdown in slab (hadsm3/hadsm3mh) models

Author Message
Profile geophi
Volunteer moderator
Send message
Joined: 7 Aug 04
Posts: 1666
Credit: 31,512,662
RAC: 55,017
Message 30789 - Posted: 3 Oct 2007, 17:45:01 UTC
Last modified: 27 Mar 2010, 1:22:01 UTC

Several crunchers have noted significant slowdowns when running a slab (hadsm3 or hadsm3mh) model, where trickle durations greatly increase, and seconds per timestep listings also increase. Following is information and recommendations concerning this problem:

Symptoms:
- Globe graphic temperature display always shows a blue (very cold) world.*
- Progress (as shown in % of model done, and timesteps increasing) is being made, but it\'s very slow.
- MUCH longer time intervals between trickles than before the slowdown first occurred. This also means that the cumulative s/TS (as seen in the globe graphic text and the trickle listing on the webpage for your model) will increase with time. You will only see this is you let it run long enough, since once it slows down, the duration between trickles becomes days instead of hours.**

Causes:
- Model instability due to parameters (see note 1 below about Intel/Windows)
- Possibly an unstable computer

Recommendations:
- If very close to end of model (like within one more trickle), perhaps allow it to run to the end, unless credits are a big deal to you
- Otherwise, manually abort the errant task
- In case it is due to hardware problems, follow recommendations in Hardware forum for hardware specific tests and checks of your PC and basic hardware maintenance to ensure your system is running cool, stable and clean.
- If overclocking, you might try to decrease or eliminate the overclock
- If you want to help moderator Iain Inglis track the locations of the point of failure, and you have a backup before the model \"froze\", you can find instructions for doing that in this thread.


* Blue/cold worlds can also occur with fast processing models and normal speed models as well. But the slow processing ones are the most troublesome since they take forever to finish.
** There are also a number of other possibilities as to why a trickle that is expected to occur near a certain time, does not, or progress is not being made. One is a looping model. This Readme talks about looping in the coupled (hadcm3) models, but much of it applies to looping in hadsm3 models as well.


If you think you may have a model like this and are not sure if it is one of these \"slow-processing iceballs\", please post a reply to this thread with the following information:


  1. A link to the model/ResultID webpage
  2. A current timestep of that model (on the globe graphic)
  3. The s/TS value (on the globe graphic. Remember, you can hit the Z key while viewing the globe and it will give you this additional text/status information.)
  4. Whether the temperature display of the globe graphic is blue.
  5. What your processor/CPU and Operating System is (i.e. Intel or AMD on Windows or Linux)
  6. Whether you are overclocking.


Note 1 - The computers that occasionally run into this problem have Intel Pentium 4 or later processors running on the Windows operating system. It is unknown at this time why some models turn into slow-processing iceworlds on this platform, and not on Macs, AMDs and any computer running Linux. Occasionally Windows/AMD will have a fast processing iceworld.

Modified Feb 21 2010 for details in red.

Profile mo.v
Volunteer moderator
Avatar
Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 10,731,610
RAC: 4,301
Message 30791 - Posted: 3 Oct 2007, 20:53:12 UTC
Last modified: 3 Oct 2007, 20:54:07 UTC

One member reported a slab model that speeded up a lot. It was discovered that this model, though apparently running, had stopped processing its climate. So if your model speeds up greatly you should also ask for advice in Geophi\'s linked thread.
____________
Cpdn news

Message boards : Number crunching : Iceworlds and Slowdown in slab (hadsm3/hadsm3mh) models


Main page · Your account · Message boards


Copyright © 2017 climateprediction.net