Posts by marmot

1) Message boards : News : New app, YAFU-64t (Message 1703)
Posted 11 Jan 2023 by marmot
Post:
This 64t is about to break 5000 CPU hours.
I'm thinking it's time to abort.


Had to restart the machine and it went back to ECM mode.

It's aborted.

Welp, at least another blue app for WUProps....
2) Message boards : Number crunching : When will you update yafu to version 2? (Message 1701)
Posted 10 Jan 2023 by marmot
Post:
Yes. The amount of time & power wasted on stuck work units is a real turn off, especial with current high power prices. You never quite know if it's stuck or still going.
Abandoning 20,30+ hrs of crunching is very wasteful.


This current 64t is at 4867 hours of CPU time, almost 7 days of run.

No credit coming and I could have been doing SRBase.
3) Message boards : News : New app, YAFU-64t (Message 1700)
Posted 10 Jan 2023 by marmot
Post:
This 64t is about to break 5000 CPU hours.
I'm thinking it's time to abort.
4) Message boards : News : New app, YAFU-64t (Message 1699)
Posted 7 Jan 2023 by marmot
Post:
The checkpoint file which is used for resum is nfs.dat. This file is created after the ecm phase and after the first round of gnfs.
Maybe the app was not yet there because of the large composites which are now handled in 64t and 128t app.

Maby you can use Windows task scheduler to schedule Windows to sleep and wake automatically:
https://www.groovypost.com/howto/schedule-wake-sleep-windows-automatically/
https://superuser.com/questions/1069944/use-windows-task-scheduler-to-put-windows-to-sleep-without-batch-file

yoyo


This 64 is into 4th day; partly because it was warmer weather and I turned down the CPU frequencies.

Someone at WUProp gave me a batch file, using boinccmd, to dual schedule BOINC pause with task scheduler.

The 2nd link you offered looks promising.
The 1st is for Windows 10 but the questioner's machine would not wake back up and made a new question here;
https://stackoverflow.com/questions/32360306/sleep-via-shortcut-causes-schedule-tasks-to-not-wake-computer

In the comments is offered to Download Sysinternals PsTools..
So I'll try that if task scheduler wake fails.

These machines run a severely stripped down Windows 8.1, with all but a few services still enabled, to reduce RAM and CPU usage by OS.
I'll have to turn the task scheduler back on!
5) Message boards : News : New app, YAFU-64t (Message 1696)
Posted 2 Jan 2023 by marmot
Post:
Our electric company has pushed out smart meters to all the customers and now is pushing new rate plans.

I took the plan that charges 4.1 cents /KWh but I have to pause electric use from 6-8am and 6-8pm (refrigerator's are on timers) and all the computing machines (which double as winter heaters) pause 4 hours a day.

One 128t completed in 11 hours, within the time where it wouldn't pause.
The 64t is pausing and staying in RAM during the peak hours and it appears to be able to complete as long as it stays in RAM paused.
Getting the servers to sleep and wake back up is not solved and BOINC does not support 2 pause periods per day (clunky dual installs.. just improve the BOINC app devs!).

But, my new CPU's showed up and I had to shut down the machine.

The 64t was in the final stage where it was just yafu.exe using all threads and slowly down-threading to 1 and then bouncing back up.

After restarting the machine, this WU is starting all over again in the ECM phase...
What happened to the checkpoint?
Why is this WU starting all over from the beginning?

I remember this being an issue 5 years ago, where a WU would start completely over if the machine rebooted.
6) Message boards : Number crunching : 16t work units not properly credited (Message 1368)
Posted 28 Feb 2019 by marmot
Post:
118000 credits for a wu ;D

https://yafu.myfirewall.org/yafu/workunit.php?wuid=3434751

thats more what i get in the last 2 days :D
why i should participate in a projekt where the credits system is totally broken.
or missed i something?!



The main BOINC forums explained that BOINC Credit New system works well on floating point but not well on integer-only calculations.
I think the oddity of Credit New is the issue here and it's out of Yoyo's hands.
7) Message boards : Number crunching : "job cache full" (Message 1367)
Posted 28 Feb 2019 by marmot
Post:
Yes, the project is configured in a way that you will have only 2 workunits on your host.
There are reasons for it.
Sometimes I change it to 3.


Yes, yes, yes, I've been working on the project for 3 years now and this is not news.

Last month yafu small sent down WU even if the cache was full.
Now it refuses.

Have to pause other high priority project manually and sit and request yafu-small manually 10- 20 times (per 32 computers, 300+ mouse clicks and an hour of my time) to get 2 tasks per machine that might be done in 3 minutes or 3 hours.

Something has changed.

I want to get 100,000 WUProps hours in on yafu-small but can't allow it to have all the cores as it's too unreliable to obtain, doesn't pay well and it's central thread is set to normal priority (not idle) so it dominates the cores and won't share cycles.
8) Message boards : Number crunching : "job cache full" (Message 1361)
Posted 27 Feb 2019 by marmot
Post:
yafu-low used to send 1 or 2 units down, now it gives the job cache full error.

Since it will never send down more than 2, and it is often empty server side, another project will always fill up the cache and no more YAFU. smalls.

Both projects are set to 49 resource share and a 3 day cache with 0 additional.

I don't want to be manually pausing one project then requesting YAFU-LOW every 2 hours.
9) Message boards : Number crunching : YAFU 4T freeze computer (exept task manager !!!! (Message 1210)
Posted 12 May 2018 by marmot
Post:

Isn't there a way to specify the process names which should be throttled?

yoyo


In TThrottle there is a way to add processes to the recognized list.
I tried using it for YAFU sub processes, but failed after an hour of attempts, so resorted to a global throttle of ThrottleStop (which was already installed and ready to go).

I just tried setting BOINC Manager to 75% maximum CPU usage and it is properly throttling gnfs-lasievexxx today as the processes are going gray for split seconds avery few seconds in process manager. (I remember that prior version of BOINC wasn't doing this but my memory is getting older now and might not be reliable).

So BOINC Mgr can throttle YAFU and protect the CPU from overheating and the OP's problem should be solved.

@marsinph, if you set BOINC to use at most 6 cores then you can only run yafu, yafu small composite and yafu-4t. If you made changes to app_config.xml to allow for 16t by setting virtual core count to 16, then 75% of 16 is 12 cores. At 12 virtual cores you are still using all 8 real cores on Yafu, Yafu composite, Yafu 4t and Yafu 8t if you haven't set specific core and thread counts for each work unit.

So if your virtual core count is 16, you need to set BOINC to 37% or change the app_config.xml (or delete it).

EDIT: Forgot this thread was about 4t only. BOINC setting to 80% CPU is working fine for a 4t WU on this laptop as I type this. Using Process Hacker to watch the WU throttling down.
10) Message boards : Number crunching : YAFU 4T freeze computer (exept task manager !!!! (Message 1207)
Posted 5 May 2018 by marmot
Post:
Hello,
it do perfect the same on my second host. https://yafu.myfirewall.org/yafu/results.php?hostid=33275
Once again Yafu take 100% non stop.
In my BAM also in my computing preference here on my account page ,
there are restriction : " use max 90% cpu time (.to reduce and prevent over heat).
Yafu ignore it. With results of host not more can do something (exept task manager.
I will not put my CPU in danger for Yafu who ignore the settings.

:

Tech PowerUp ThrottleStop is one solution to globally protect CPU's from overheating if TThrottle and BOINC MGR fail on Windows.


@yoyo_rkn, I understand what they are talking about now.
I have seen the same problem. I even installed TThrottle to idle WU's in order to protect my laptop from over heating and it will throttle down yafu_wrapper but the sub processes do not receive a throttle control message and continue at 100% and overheat the CPU.
It could be that TThrottle doesn't recognize the sub processes started by wrapper as being BOINC processes. But BOINC.exe also refuses to throttle them YAFU sub processes and this is what marsinph is complaining about.

Maybe it's just a global variable that's not being exported to the OS environment that TThrottle and BOINC.exe need to see in order to throttle down the ecm/gnfs-lasieve sub procs?

Protecting machines from overheating by limiting BOINC to 80% max CPU is legitimate need for users.

To recreate this, use your test machine and install TThrottle, tell it to keep the CPU temp under 70C, then watch it's behavior on YAFU and other projects. For LHC Theory or ATLAS, it will lower the VM's CPU usage till the CPU temp drops to 70C. For sixtrack.exe, it throttles the executable using OS commands. It throttles yafu_wrapper but not ECM/GNFS.
Also, see what BOINC.exe does with YAFU when telling it to use at most 80% CPU.
11) Message boards : Number crunching : How do I get YAFU to stop sending me tasks that use up 100% of my cores? (Message 1202)
Posted 20 Apr 2018 by marmot
Post:

This is why I reverted back to the 4t and 8t processes. At least they run no more than what they claim allowing BOINC to assign those idle cores to different projects.


If you watch those tasks, you will see they do the same thing. The GNFS or ECM sub tasks exit one by one as they finish their data set and eventually the 4t or 8t will also only use 1 core for some minutes. The 32 and 16t are just more noticeable as the data sets are larger and they count down from larger thread counts of 16 or 32 threads to finally reach 1 thread then a restart with a new subset.

You can set the thread count on a 16t to 16 threads and max_concurrent to 1 in App_config.xml and assure that 48 cores are left for other projects.
12) Message boards : Number crunching : YAFU 4T freeze computer (exept task manager !!!! (Message 1201)
Posted 20 Apr 2018 by marmot
Post:
If yafuwrapper_26014.......exe is taking up 100% CPU then it's probably fighting with the firewall/security application.

Make sure the firewall or other security software isn't blocking the wrapper executable.

Open up the Boinc\Data directory in your firewall/security software.

The wrapper code just manages the yafu.exe, ecm.exe and gnfsxxx.exe tasks and shouldn't use more than 0.2% CPU.
13) Message boards : Number crunching : What's the largest CPU cycles yet? (Message 1200)
Posted 20 Apr 2018 by marmot
Post:
Bet you can't beat this one.
You'd think it'd get more credit than 10k.

Task: 2103181
Sent: 7 Apr 2018, 15:23:16 UTC
Reported: 18 Apr 2018, 3:00:48 UTC
Status: Completed and validated
Run time (sec): 905,852.00
CPU time (sec): 22,161,520.00
Credit: 10,285.52
Application: YAFU-32t v134.05 (32t) windows_x86_64


I'm done with 32t WU since the warm weather is coming now.
14) Message boards : Number crunching : What's the largest CPU cycles yet? (Message 1179)
Posted 23 Mar 2018 by marmot
Post:
This one is 80% longer than my last record holder.

Task: 1943331
Sent: 20 Mar 2018, 9:45:57 UTC
Reported: 23 Mar 2018, 8:01:38 UTC
Status: Completed and validated
Run time (sec): 252,941.00
CPU time (sec): 9,559,034.00
Credit: 36,097.01
Application: YAFU-32t v134.05 (32t) windows_x86_64
15) Message boards : Number crunching : What's the largest CPU cycles yet? (Message 1178)
Posted 20 Mar 2018 by marmot
Post:
This is my record holder:
Task: 1981055
Sent: 17 Mar 2018, 7:30:14 UTC
Reported: 20 Mar 2018, 9:45:57 UTC
Status: Completed and validated
Run time (sec): 267,343.00
CPU time (sec): 5,230,671.00
Credit: 37,854.22
Application: YAFU-32t v134.05 (32t) windows_x86_64
16) Message boards : Number crunching : More than 2 WU's? (Message 1141)
Posted 23 Feb 2018 by marmot
Post:
The problem I had was that BOINC decided to work on two WU's for 2 days then suspend one of them and swap in a new WU on the 3rd day and run it for a couple days.

I told BOINC to swap WU's only every 9999 minutes but it either completely ignores that number and swaps sometimes or it's flawed calculations decide the new WU won't meet it's deadline and swaps it in for priority run.

Now that there is a steady supply of 32t, I'm going to set preferences for 32t to d/l only 1 WU so BOINC can't possible make a swap and I'm not getting the "not highest priority project" d/l failure on shared projects.
17) Message boards : Number crunching : More than 2 WU's? (Message 1136)
Posted 21 Feb 2018 by marmot
Post:
Some of my machines are now getting 3 and 5 WU to d/l.

You made a permanent change or is this an experiment?

The 2 WU limit actually is useful in getting YAFU to share cores with other projects as it avoids the dreaded 'not highest priority' work fetch error when BOINC sees WU's near their deadline and doesn't understand that app_config is limiting projects to 4 of 16 or maybe 4 of 48 cores.

This is a good change for people wanting to run 4x 4t on their 16 core thread ripper.

Just need to change the work preferences to limit to 2 WU if we want back the old limit, right?
18) Message boards : News : Yafu back online (Message 1049)
Posted 12 Aug 2017 by marmot
Post:
Welcome back!
19) Questions and Answers : Wish list : Was There Ever A Badges Thread (Message 991)
Posted 8 May 2017 by marmot
Post:
I agree.
People love their badges and if they don't get any they move on...
20) Message boards : Number crunching : 0% WU... (Message 990)
Posted 8 May 2017 by marmot
Post:
Hi

I have this WU that has now over 25 hours of computation, it is using CPU and processes are regularly started / stopped on the 4 cores, the logs in slot directory all show that there is growing activity done by the WU (factor.log, out file, etc) however the progress % is 0, there is no remaining time displayed (---) and the deadline is tomorrow morning...

Is it normal ? what should I do ?

Thanks.


The first 2 WU any machine receives appear to be test WU's to verify the machine's capability to run YAFU and so they complete within a minute or three. A few WU will end in under 20 minutes and I've seen one WU go for 11 days. The 8T WU can typically run for 3 days depending on the CPU.


Next 20




Datenschutz / Privacy Copyright © 2011-2024 Rechenkraft.net e.V. & yoyo