Message boards :
Number crunching :
Run Time Estimates
Message board moderation
Author | Message |
---|---|
Send message Joined: 31 Dec 07 Posts: 1152 Credit: 22,077,407 RAC: 1,835 |
I realize that the Admins made some sort of adjustment to the time estimates for the hadam3pm2�s, but, would it effect the time remaining estimates of the other types of models. I ask because the estimated time for an unstarted hadam3p_anz that formally read 211 hours now suddenly stands at 185. Other types of models are also changed. Hadam3p_eu models have declined from 81 hours to 73. The original run time estimates were very accurate so now they will be low. |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
A run time estimate is one of many bits of info supplied to peoples computers when they get a new data set. (Another is the url's to upload the data to.) All models, and thus all model types, have their own set of estimates, as supplied from a template for that model type. (Makes it easy to set up a value for many data sets.) The estimate for the MOSES II model, (and I use this name because it's easier than remembering the string of characters for the program), was for a 120 year run, instead of the 10 year run that was being issued. If you're now having problems with another type of model, you'll have to start again with details of what's wrong with it/them. (Perhaps you've finished a model, and that's changed the FPU contention for the remaining models?) |
Send message Joined: 1 Jan 07 Posts: 943 Credit: 34,262,558 RAC: 10,761 |
I realize that the Admins made some sort of adjustment to the time estimates for the hadam3pm2�s, but, would it effect the time remaining estimates of the other types of models. I ask because the estimated time for an unstarted hadam3p_anz that formally read 211 hours now suddenly stands at 185. Other types of models are also changed. Hadam3p_eu models have declined from 81 hours to 73. The original run time estimates were very accurate so now they will be low. This is absolutely normal for BOINC. BOINC itself keeps track of the accuracy of runtime estimates, and will have noticed that the MOSES II "ended early", according to the original estimate. BOINC's rather simple-minded assumption is that all tasks behave the same - so if this one finished early, the next one is likely to as well. It starts adjusting something called the 'Duration Correction Factor' (DCF) - very gradually, but eventually DCF would ensure that the displayed estimates were correct on your machine, even if you ran a consistent series of MOSES II tasks with the faulty estimates. But BOINC only has one DCF, and applies it to all types of model (that will change after the server upgrade), so the 'correction' is applied to the already-correct estimates too. It'll be adjusted back the other way when this model finishes. |
©2024 climateprediction.net