 |
|
Message boards : Number crunching : Cancel high priority run now or later?
Message board moderation
| Author | Message |
|---|---|
 old_user280873 Send message Joined: 18 Feb 06 Posts: 17 Credit: 1,769,142 RAC: 0 |
My pc is working on an Ocean 6.07 model with high priority. The calculation must be done before Aug 4th 2011. My pc will not running all the time, so I can't finish on time. What is the best thing to do. Cancel now or cancel in August? Leendert. |
|
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
How about option C)? Don't cancel at all. As has been posted many times, there IS no deadline with cpdn! But the BOINC software insists on one, (because of how other projects work), so a fake one is used. The data will still be accepted long after the "deadline" expires. The reason that it's so short this time, is that this first batch is needed quickly to use as input to the next stage. The researchers are on a tight time line, so they're relying on those computers that are running fast 24/7 to provide as many as possible as soon as possible. The slower returns will be used later. And BOINC will gradually work out that it's not as bad as originally thought, and get out of 'panic mode'. But that will take a week or so. A thread by the researchers is here on our php board. There are also some threads on this board about the models. Backups: Here |
|
Iain Inglis Send message Joined: 16 Jan 10 Posts: 1084 Credit: 7,808,726 RAC: 5,192 |
Carry on: the data will still be accepted after the deadline. The 'time-to-completion' estimate may also be too high in the early stages of the model and will come down as the model progresses: you may therefore find that the model does in fact have enough time to finish. [Edit: Les got there first.] |
|
Send message Joined: 10 Jul 11 Posts: 4 Credit: 18,510 RAC: 0 |
Isn't any deadline? So I get EXACTLY the same credit, even if I upload a result the same day I downloded it? |
|
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
With this project, credit is given for trickle_up files returned. These files are created at regular intervals, which differs for each type of model, and they contain some of the result data. The rest is contained in zip files, also created at regular intervals. No trickles, no credit. And the shortest model at present takes about 3 days on the fastest processor, running 24/7, with no other projects competing for time. So it's impossible to finish a climate model in just one day. Backups: Here |
|
Send message Joined: 10 Jul 11 Posts: 4 Credit: 18,510 RAC: 0 |
How much credits will I earn then? |
|
Send message Joined: 3 Nov 10 Posts: 39 Credit: 2,494,427 RAC: 0 |
markus: there are 40 trickles - each trickle is 311 credits; therefore, you will receive about 12,440 credits for the completed workunit... |
 Send message Joined: 28 Nov 06 Posts: 89 Credit: 11,986,335 RAC: 2,269 |
How about option C)? Don't cancel at all. An excellent question from Leendert... I want to add something. For Ocean models we have: minimum quorum - 1; initial replication - 2; max # of error/total/success tasks - 3, 3, 3. So, if some of 2 hosts will miss the "deadline", 3rd extra-task may or will be sent. Questions... 1. This extra-task, if it will be sent? Is it not a CPU-time waste? ;-) 2. My behavior, if I got this 3rd task? Situation - the 2nd host, when my host got 3rd task, was approaching to end of processing and finished it's job few days after it's "deadline", my host has a small progress at that moment. My intuition says - if I don't want to abuse my CPU for useless job, I need to cancel this extra-task. Am I at least partially correct? Some unpleasant details... 1. I am multi-project participant. CPDN was always a friendly and comfortable project. Now it is a conflicting project, if running Ocean models, because it is blocking another projects, especially when BOINC is upgraded to newest versions. 2. [IMHO] - I don't understand sometimes, how IQ of BOINC is evolving, because at least since versions 6.2.x it is using more and more strange arithmetic - 2x2 is no more 4. [/IMHO] Effects of this are - the high priority is assigned to the task without some explainable reason or, much worse, the task is not running yet in high priority, when it is necessary already. As result of this - the Ocean tasks are missing their "deadlines" on older machines - an example... And I have more... 
|
|
Send message Joined: 10 Jul 11 Posts: 4 Credit: 18,510 RAC: 0 |
So that means no tasks gonna be deleted? |
|
Send message Joined: 10 Jul 11 Posts: 4 Credit: 18,510 RAC: 0 |
So that means no tasks gonna be deleted? |
|
Send message Joined: 28 Nov 06 Posts: 89 Credit: 11,986,335 RAC: 2,269 |
So that means no tasks gonna be deleted? 1. Your behavior will be correct here if and only if You will finish all of tasks, because the combination of parameters may be unique for every task of the same WU. 2. CPDN servers are accepting everything - every trickle Your machines will send to them.
|
©2024 cpdn.org
