𝕏

New version 0.04 of DENIS-Fiber_beta with checkpoints

Message boards : Number crunching : New version 0.04 of DENIS-Fiber_beta with checkpoints
Message board moderation

To post messages, you must log in.

AuthorMessage
Crystal Pellet

Send message
Joined: 16 Jul 15
Posts: 15
Credit: 6,451,762
RAC: 3,202
Message 3109 - Posted: 26 May 2025, 10:00:40 UTC
Last modified: 26 May 2025, 10:12:00 UTC

The released version 0.04 from this morning seems to have working checkpoints - Great!

But the tasks also have very long estimated runtimes . . . with an extended deadline up to 22.1 days.
ID: 3109 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3111 - Posted: 26 May 2025, 10:30:57 UTC - in response to Message 3109.  
Last modified: 26 May 2025, 10:36:04 UTC

Hmm, I didn't manage to get any tasks, even though DENIS is set to 99,99% resource share. Resetting or manually dropping and re-adding the project didn't help either, I couldn't force BOINC to download the new application.

Edit: I think I found the explanation. There were only 2000 tasks sent out and none of my rigs was actively bugging the project scheduler in the few seconds that they were available.


ID: 3111 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile ivanbisimbrero
Project developer
Project scientist

Send message
Joined: 21 Nov 24
Posts: 21
Credit: 63,260
RAC: 774
Message 3112 - Posted: 26 May 2025, 14:02:14 UTC - in response to Message 3109.  

Hello!

That the work packages of this version take longer than usual is completely deliberate.

For a first test we have launched 1000 tasks, with a simulation time per configuration of about 100 seconds (until now, they were being 5 to 10 seconds).

This was done to test the effectiveness of the checkpoints, so that we can get more realistic simulation results thanks to the checkpoints.

We will keep you informed. Thank you very much.

Ivan.
ID: 3112 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile rilian
Avatar

Send message
Joined: 21 May 15
Posts: 31
Credit: 869,420
RAC: 10,201
Message 3114 - Posted: 26 May 2025, 19:46:52 UTC - in response to Message 3112.  
Last modified: 26 May 2025, 19:48:21 UTC

That the work packages of this version take longer than usual is completely deliberate.

This was done to test the effectiveness of the checkpoints, so that we can get more realistic simulation results thanks to the checkpoints.

Ivan.

Hi Ivan, do you require us to suspend work unit, resume, quit boinc manager, launch boinc manager for this test ?

PS: on my tasks i now see
CPU time = 09:13:00
CPU time since last checkpoint = 00:00:13 ...
--
I crunch for Ukraine

ID: 3114 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Keith Myers
Avatar

Send message
Joined: 19 Jun 24
Posts: 5
Credit: 186,233
RAC: 0
Message 3119 - Posted: 27 May 2025, 4:34:10 UTC - in response to Message 3114.  

You can do all that on your own. Easy enough to check the checkpoint file in the slot of a running task and see whether it checkpoints properly under all stopping/suspending conditions.
ID: 3119 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3120 - Posted: 27 May 2025, 8:28:03 UTC
Last modified: 27 May 2025, 8:43:02 UTC

New work incoming! Grab it while it's hot!



Estimated runtime is about 20x longer than for v0.03
ID: 3120 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile ivanbisimbrero
Project developer
Project scientist

Send message
Joined: 21 Nov 24
Posts: 21
Credit: 63,260
RAC: 774
Message 3122 - Posted: 27 May 2025, 9:29:27 UTC - in response to Message 3120.  

Hi!

Estimated runtime is about 20x longer than for v0.03


Yes! But it's not a performance issue, is due to the simulation time that we're sending via config (now 100 seconds vs in previous version, that we're sending simulations of 5-10 seconds).

Iván.
ID: 3122 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3123 - Posted: 27 May 2025, 10:05:52 UTC - in response to Message 3122.  

I know, I was just reporting the order of magnitude so you guys can compare it to other results and estimates :)

I have a miniPC dedicated to running BOINC that has 8 identical cores (Ryzen 7 5825U, dynamically throttled to 70 C) so it's good for benchmarking runtimes. From about 0:48 with v0.03 the current runtime estimates are 17:12.

My personal laptop right now estimates about 26 hours for two of the running tasks, and 55-59 hours for two different tasks. It has a Core 5 Ultra 125H with two very different core types, all limited to base speed (no turbo at all). I've seen a similar 2x difference in runtime with other projects as well, so it's safe to assume that everything is running normally, with a big efficiency difference between the two core types.

3 other laptops all have single-type cores (i5-6200U, i7-5600U, i7-1165G7) and current runtime estimates also range between 20-25x compared to v0.03.
ID: 3123 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile rilian
Avatar

Send message
Joined: 21 May 15
Posts: 31
Credit: 869,420
RAC: 10,201
Message 3128 - Posted: 27 May 2025, 15:47:01 UTC - in response to Message 3123.  

I have "request tasks to checkpoint at most" setting set to 60 seconds, and i observe it checkpoints about every 80-90 sec
--
I crunch for Ukraine

ID: 3128 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3134 - Posted: 28 May 2025, 6:09:12 UTC

Granted credits are in line with expectations, from 20-25 up to 450-500.
ID: 3134 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3156 - Posted: 2 Jun 2025, 15:24:09 UTC
Last modified: 2 Jun 2025, 15:25:57 UTC

There was a new batch of 1000 WUs/2000 tasks, but because of its small size it pretty much disappeared instantly. I only managed to grab 16 of them on my dedicated MiniPC.

On a side note, I got a resent task for a WU in which both initial tasks came back inconclusive. Anyone else noticed such resends? Would be interesting if the team told us what causes this phenomenon.
ID: 3156 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Lanius collurio

Send message
Joined: 5 Apr 25
Posts: 53
Credit: 255,016
RAC: 9,143
Message 3157 - Posted: 2 Jun 2025, 15:24:22 UTC
Last modified: 2 Jun 2025, 15:24:56 UTC

And as always, double post when I'm on my phone... Please delete... Thx!
ID: 3157 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jean-David Beyer

Send message
Joined: 6 Mar 23
Posts: 71
Credit: 2,442,785
RAC: 1,309
Message 3158 - Posted: 2 Jun 2025, 18:54:38 UTC - in response to Message 3156.  

There was a new batch of 1000 WUs/2000 tasks, but because of its small size it pretty much disappeared instantly. I only managed to grab 16 of them on my dedicated MiniPC.


It was so small I did not get any.

On a side note, I got a resent task for a WU in which both initial tasks came back inconclusive. Anyone else noticed such resends? Would be interesting if the team told us what causes this phenomenon.


No resends either.
ID: 3158 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : New version 0.04 of DENIS-Fiber_beta with checkpoints