RAC going down

[Rockinfroggi]

Can anyone tell me, does CPDN update it's stats less frequently than other projects? I restarted it on 3 of my systems a few days ago and my RAC is going down instead of up.
I know they are large WU's (deadline being Dec this year) but I thought they were supposed to trick feed your credits?


Gary.

[DJH@GB-Ro]

Hi Rockinfroggi,

Although you do get trickles, these only happen every 25,920 timesteps (1 model year).

I looked at your results and only one of your computers has sent trickles so far, and it has sent two, with around 3.25 sec/TS (takes 3.25 second for each of the timesteps).

Click: your top computer for CPDN

At that speed, 25,920 x 3.25 = It will take 84,240 seconds (or 23.4 hours) before each trickle is sent. So if you are sharing CPU time over multiple projects, you may only get trickles every few days or even longer depending how much CPDN you crunch in a day.

There is still a delay between trickles received and credit granted. Your credit will double to 518.4 very soon as your second trickle has been sent, thus sending up your RAC.

So nothing to worry about, it may seem like a long time between getting credit, but when you do it will be 259.2 each time.

[DJH@GB-Ro]

I also noticed your P4 3.0GHz with HT's models are 'over':

http://climateapps2.oucs.ox.ac.uk/cpdnb ... tid=526787 (if you click results, you'll see)

It doesn't say error, so were they manually aborted?

[Rockinfroggi]

I also noticed your P4 3.0GHz with HT's models are 'over':

http://climateapps2.oucs.ox.ac.uk/cpdnb ... tid=526787 (if you click results, you'll see)

It doesn't say error, so were they manually aborted?

Thanks for looking into that for me, I'll soon get used to all this and stop asking silly questions.

As for my 3.0 HT system, I suspended that along with all the others when we did the mini crunch, I've just not resumed it on that system.

I thought as the deadlines on these WU's were so long it would be ok to share resources with other projects but would it be better to devote one system to CPDN alone and use the others for other projects or is it a case of six of one half a dozen of the other?

Thanks again for taking the time to look at this for me

Gary.

[DJH@GB-Ro]

I thought as the deadlines on these WU's were so long it would be ok to share resources with other projects but would it be better to devote one system to CPDN alone and use the others for other projects or is it a case of six of one half a dozen of the other?

If the amount of time spent in total was equal whether is was spread over the 4 machines, or just one, then the credit you receive and work done will be equal. The difference however, is if you just had one machine fully on CPDN, the models would get completed much more quickly. So it is a choice of having less models but completing them more quickly, or many which may take longer.

Looking at your results, only one machine has completed model years, having done 2 of 160 years. The more recent year started on 17/01/07 and finished today 06/02/07, taking 20 days. At that rate, it would not finish by the deadline. But all data is used, whether the model finishes or not.

If it was me, I'd definitely devote one machine to it, and it would be the HT machine as HT (or dual core) do 2 models simultaeneously which is more efficient that just having one.

Hope this helps

P.S. If you want more detail, these sort of issues are discussed regularly on the CPDN mesage boards with opinions of people who have more knowledge than me.

[DJH@GB-Ro]

Thanks again for taking the time to look at this for me

No problem :thumbup: Glad to help (and do my part as CPDN Cheerleader! )

[Rockinfroggi]

I thought as the deadlines on these WU's were so long it would be ok to share resources with other projects but would it be better to devote one system to CPDN alone and use the others for other projects or is it a case of six of one half a dozen of the other?

If the amount of time spent in total was equal whether is was spread over the 4 machines, or just one, then the credit you receive and work done will be equal. The difference however, is if you just had one machine fully on CPDN, the models would get completed much more quickly. So it is a choice of having less models but completing them more quickly, or many which may take longer.

Looking at your results, only one machine has completed model years, having done 2 of 160 years. The more recent year started on 17/01/07 and finished today 06/02/07, taking 20 days. At that rate, it would not finish by the deadline. But all data is used, whether the model finishes or not.

If it was me, I'd definitely devote one machine to it, and it would be the HT machine as HT (or dual core) do 2 models simultaeneously which is more efficient that just having one.

Hope this helps

P.S. If you want more detail, these sort of issues are discussed regularly on the CPDN mesage boards with opinions of people who have more knowledge than me.

Thanks for the update, I'll give it some thought but I guess it would mean aborting the partially completed CPDN WU's on my other systems and start afresh on the HT system, though I would probably suspend it for any mini crunches.
I'll hold fire for a day or two and see if anyone else has anything to add then may well take your advice.


Gary.