Message boards : Number crunching : AE_2 / AEty_2 Units?
Author | Message |
---|---|
The-Real-Link Send message Joined: 27 Dec 10 Posts: 6 Credit: 2,676,652 RAC: 0 |
Greetings, Just noticing that I'm getting some of these newer units (I'm guessing that the project is at a new stage or trying something else). The main question I had is whereas a normal Rosetta unit completes for me in a handful of hours, these AE units seem to be around 10-12 hours each. Is that normal? If this is a duplicate thread, sorry. |
Murasaki Send message Joined: 20 Apr 06 Posts: 303 Credit: 511,418 RAC: 0 |
Rosetta work units aim to run to match your preferred run-time. If you have a run-time of 7 hours BOINC will keep running variations of the experiment (a "decoy") for the task until it gets close to the 7 hours. At the end of a decoy BOINC will take a look at the average time recent decoys have taken and decide if there is enough time to run another; e.g. if 5 hours have passed and decoys are only averaging 1 hour each then BOINC will try another decoy, but if 5 hours have passed and decoys are averaging 2 hours each then BOINC will see you will exceed the 7 hour run-time and stop at that 5 hour mark. Sometimes the last decoy may be more complex or BOINC miscalculates the average so you could exceed your preferred run-time by up to 4 hours (at 4 hours beyond the preferred run-time the "watchdog" activates and terminates the task in case the delay was caused by an error). These factors can lead to a great variety in the times that the different tasks run for. However, taking a look at one of your recent tasks the output report says you have "cpu_run_time_pref: 43200". That means BOINC thinks your preferred run-time is 43,200 seconds or 12 hours. If you don't want Rosetta tasks to run for that long you may want to check your preferences. |
The-Real-Link Send message Joined: 27 Dec 10 Posts: 6 Credit: 2,676,652 RAC: 0 |
Rosetta work units aim to run to match your preferred run-time. If you have a run-time of 7 hours BOINC will keep running variations of the experiment (a "decoy") for the task until it gets close to the 7 hours. At the end of a decoy BOINC will take a look at the average time recent decoys have taken and decide if there is enough time to run another; e.g. if 5 hours have passed and decoys are only averaging 1 hour each then BOINC will try another decoy, but if 5 hours have passed and decoys are averaging 2 hours each then BOINC will see you will exceed the 7 hour run-time and stop at that 5 hour mark. Sometimes the last decoy may be more complex or BOINC miscalculates the average so you could exceed your preferred run-time by up to 4 hours (at 4 hours beyond the preferred run-time the "watchdog" activates and terminates the task in case the delay was caused by an error). These factors can lead to a great variety in the times that the different tasks run for. Thanks for taking the time to go into detail about all that. In part, I found for some reason I was experimenting and had moved my CPU cores back up to 16 so both Rosetta and Milky Way were both "fighting" over CPU time and slowing both down. As soon as I moved my settings back, all the lengthy pending tasks suddenly finished at once, though a few were already near 12 hours anyway. I guess I will have to go back in my preferences and take a look at that. So am I correct in understanding that I should set that run time pref down to say, an hour? That way I'm not losting computational time or credit? |
The-Real-Link Send message Joined: 27 Dec 10 Posts: 6 Credit: 2,676,652 RAC: 0 |
Yep, you're right - I had Rosetta-specific settings for CPU-time set to 12 hours. Perhaps I just thought that that way it'd work on crunching Rosetta for that long before it'd switch to a Milky Way project (which I wouldn't want it to do). I didn't realize it'd make each project attempt a 12-hour run or scale to a bigger unit that'd take 12 hours. I reset it to 3 as that's been a good turnaround for my usage. I hope the 12 hour units do give a lot of credit for the time spent though! ;) |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
I like how your 16-core server beats all of my PCs combined. I take yours runs 24/7. |
The-Real-Link Send message Joined: 27 Dec 10 Posts: 6 Credit: 2,676,652 RAC: 0 |
Hahah um, thanks? XD Nah I just wanted to rebuild this system recently into something nice - first time for me doing a lot of this stuff to be honest. I try to run Rosetta as much as I can but depends on what I'm doing each day. I do a lot of photography and some video work too so it's not always crunching science. |
Message boards :
Number crunching :
AE_2 / AEty_2 Units?
©2024 University of Washington
https://www.bakerlab.org