Message boards : News : Scheduler update for more accurate job cache
Author | Message |
---|---|
dekim Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 20 Jan 06 Posts: 250 Credit: 543,579 RAC: 0 |
We updated the scheduler to use the optional user defined cpu run time preference for the estimated run time. Hopefully this should produce a more accurate job cache. Please post issues regarding this updae in the discussion. |
xotwod Send message Joined: 30 Mar 20 Posts: 3 Credit: 6,084 RAC: 0 |
I just got 159 Ralph@home tasks which estimate that they will take ~1 hour which does not seem possible. I'm almost certain I won't be able to meet the deadline - also the deadline shown on the RALPH website is different than what I see in my BOINC manager. |
dekim Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 20 Jan 06 Posts: 250 Credit: 543,579 RAC: 0 |
I just got 159 Ralph@home tasks which estimate that they will take ~1 hour which does not seem possible. I'm almost certain I won't be able to meet the deadline - also the deadline shown on the RALPH website is different than what I see in my BOINC manager. That doesn't sound good. How many cpu's? What do the website and manager show for deadlines? Maybe I need to reduce the duration calculation a bit to hopefully prevent this from happening. |
xotwod Send message Joined: 30 Mar 20 Posts: 3 Credit: 6,084 RAC: 0 |
https://ralph.bakerlab.org/show_host_detail.php?hostid=45448 is the computer, so literally just a AMD Ryzen 5 3600X 6-Core Processor (12 processors) Application Rosetta 4.20 Name rb_04_24_22842_22255__t000__0_C1_SAVE_ALL_OUT_IGNORE_THE_REST_39_11723 State Task suspended by user Received 2020-05-02 6:10:39 PM Report deadline 2020-05-05 6:10:40 PM Estimated computation size 40,000 GFLOPs Executable rosetta_4.20_windows_x86_64.exe is shown in BOINC manager for that task, all of the tasks I have show a deadline of 11 May 2020, 22:10:40 UTC here : https://ralph.bakerlab.org/results.php?hostid=45448&offset=100&show_names=0&state=1&appid= This behaviour is consistent across all 159 tasks currently in progress on my account. |
dekim Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 20 Jan 06 Posts: 250 Credit: 543,579 RAC: 0 |
I figured out the issue causing the discrepancy. It should be fixed now with new work units. I previously added a "report_grace_period" param in the server config a while back that added 6 days to the normal 3 day deadline and that longer deadline is what is showing on the web site. I took that config param out so now there is no longer a grace period. Our R@h project does not have a grace period set in the config. How many concurrent jobs are running on your host? 6? Assuming all jobs run for 1 hour, how many will not make the 3 day deadline? I can add a factor that increases the job duration which will produce a smaller cache, if necessary to help satisfy the deadlines. |
xotwod Send message Joined: 30 Mar 20 Posts: 3 Credit: 6,084 RAC: 0 |
12 jobs can run concurrently, but after running for half an hour a task says it is around 3% complete, it's submitting after an hour so I guess sure they will all meet the deadline. I guess my question is now why would tasks submit at around 6% complete? |
dekim Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 20 Jan 06 Posts: 250 Credit: 543,579 RAC: 0 |
The BOINC client progress and remaining time estimates can be off and irregular. You can ignore those estimates and assume the jobs will run close to the cpu run time preference which defaults to 1 hour on Ralph and 8 hours on R@h. I updated the title of this news thread since it was a bit misleading initially. A run time estimate (job duration) is calculated within the scheduler code to determine how many jobs to send to a host (and thus how many jobs to cache). This estimate is now based on the user run time preference. This change does not affect the progress and remaining time estimates displayed by the client. These values can be off and irregular due to the random nature of the modeling algorithms and checkpoint frequency, and can be ignored. |
Michael E.@ team Carl Sagan Send message Joined: 29 Apr 20 Posts: 2 Credit: 19,751 RAC: 0 |
I am seeing a number of tasks sent 1-May that within BOINC manager have a Deadline of May 4 but when viewed in the web-based account information, it says 10-May. The web interface for me is: https://ralph.bakerlab.org/results.php?userid=59270&offset=0&show_names=0&state=0&appid= A screen cap of the tasks in the BOINC Advanced view is available at: http://www.wingnaprayer.golf/ralph@home-snap3.png I do not understand the inconsistency. I understand from working with Rosetta folks that there is some software measurement once your device processes a dozen or so tasks. Until that time, my feedback is to limit the number of tasks and/or double the estimated time - see my feedback starting at https://boinc.bakerlab.org/rosetta/forum_thread.php?id=6893&postid=95513#95513 and helpful replies. I will likely need to only rum ralph@home to try to enable most of the tasks to complete if the actual deadline is May 4. Mike[/img] |
dekim Volunteer moderator Project administrator Project developer Project scientist Send message Joined: 20 Jan 06 Posts: 250 Credit: 543,579 RAC: 0 |
Michael, please see my post below. This discrepancy was due to a server configuration that I recently removed which should fix the discrepancy for new jobs. The deadlines should be 3 days and should match what is on the website for new jobs. |
Michael E.@ team Carl Sagan Send message Joined: 29 Apr 20 Posts: 2 Credit: 19,751 RAC: 0 |
Thank you dekim! I did have to abort some 3 tasks because they would not finish in time. |
Dotsch Send message Joined: 4 Mar 06 Posts: 12 Credit: 13,725 RAC: 0 |
The 14 task my system with 1 hour estimated run time was from 0,75 to 1,4 hours, with one exception for result ID 5075666 which ran 4 1/2 hours. |
Brian Nixon Send message Joined: 14 Apr 20 Posts: 5 Credit: 5,933 RAC: 0 |
Does this rely on BOINC being able to determine a task duration correction factor to get the estimates right for non-default preferences? If so, I suspect you need to remove <dont_use_dcf/>from the project settings for that to be calculated and applied. |
Brian Nixon Send message Joined: 14 Apr 20 Posts: 5 Credit: 5,933 RAC: 0 |
(Reports of estimates being off, plus armchair analysis, over on the Rosetta forums: Why are my 'Remaining' time estimates so far off?) |
fpgamall Send message Joined: 3 Sep 20 Posts: 9 Credit: 0 RAC: 0 |
https://ralph.bakerlab.org/show_host_detail.php?hostid=45448 is the computer, so literally just a AMD Ryzen 5 3600X 6-Core Processor (12 processors) https://www.fpgamall.com[/url] |
intcodecor Send message Joined: 27 Oct 22 Posts: 10 Credit: 0 RAC: 0 |
Welcome to my website?https://www.intcodecor.com/ Intco Decor One-stop Home Improvement Solution: 3D Wall Panels, Wall Cladding, Skirting Boards, Wall Moldings, Crown Molding, and Decking. |
intcodecor Send message Joined: 27 Oct 22 Posts: 10 Credit: 0 RAC: 0 |
Welcome to my website?https://www.intcodecor.com/ Intco Decor One-stop Home Improvement Solution: 3D Wall Panels, Wall Cladding, Skirting Boards, Wall Moldings, Crown Molding, and Decking. |
GNHUSTGames Send message Joined: 16 Nov 23 Posts: 2 Credit: 0 RAC: 0 |
https://gnhustgames.github.io/ is a vibrant online platform dedicated to providing a diverse collection of entertaining and accessible games for users of all ages. Our website offers a wide array of unblocked games, ensuring that visitors can enjoy an extensive selection without any restrictions. |
GNHUSTGames Send message Joined: 16 Nov 23 Posts: 2 Credit: 0 RAC: 0 |
https://gnhustgames.github.io/ is a vibrant online platform dedicated to providing a diverse collection of entertaining and accessible games for users of all ages. Our website offers a wide array of unblocked games, ensuring that visitors can enjoy an extensive selection without any restrictions. |
Message boards :
News :
Scheduler update for more accurate job cache
©2024 University of Washington
http://www.bakerlab.org