PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Problems and Help : My BOINC went haywire! (Message 134590)
Posted 392 days ago by Profile Tuna Ertemalp
Have you seen this thread?

<plan_class>mt</plan_class>


Thank you!!!! I just skimmed it. Seems that is the relevant info. I will dig deeper and make the needed changes.

Tuna
2) Message boards : Problems and Help : My BOINC went haywire! (Message 134578)
Posted 392 days ago by Profile Tuna Ertemalp
Interestingly, I am seeing this in the BOINC log file when doing an UPDATE on primegrid:

11/5/2019 5:58:41 AM | PrimeGrid | Entry in app_config.xml for app 'llr321', plan class '' doesn't match any app versions

I believe this is a response returned by Primegrid to BOINCmgr. I wonder what it is complaining about... I also noticed on http://www.primegrid.com/apps.php that all LLR applications got updated recently on Sep 3rd. I wonder if any code changes resulted in some new or buggy behavior that is not the same as before regarding the reading/using of app_config. Maybe there is some info buried in the message boards about it but, as I said, I don't keep track of my BOINC stuff as closely as I used to a decade ago...

Tuna

PS: This log entry is not unique to llr321. I get it for others, as well, and I don't ever remember seeing them: "11/5/2019 6:16:26 AM | PrimeGrid | Entry in app_config.xml for app 'llrPSP', plan class '' doesn't match any app versions"
3) Message boards : Problems and Help : My BOINC went haywire! (Message 134576)
Posted 392 days ago by Profile Tuna Ertemalp
The host itself is set up as a generic host that can run any job for any project, as are all of my BOINC hosts. Recently I changed the stronger hosts (many cores, multiple high-end GPUs) to do DYFL tasks on the GPUs and the 321Sieve on the CPUs by simply setting BOINCMgr to NoNewTasks for all the other projects those hosts serve, and setting the PrimeGrid preferences for those hosts to give me only DYFL (to keep the GPU busy) and 321Sieve tasks (to keep the CPUs busy), plus "give me anything when no other jobs left" (to not idle anything for no good reason). There were no other changes to any of the (max)thread settings. Quite frankly, I just keep building big loaded machines as new GPUs/CPUs come out, and throw them at all BOINC projects/tasks, and leave them running for months and months without any change; I am not a micro-manager of my BOINC tasks nor my hosts... :)

In any case, I just changed app_config on one such machine from:

<app> <name>llr321</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llr321</app_name> <cmdline>-t 8</cmdline> <avg_ncpus>8</avg_ncpus> </app_version>


to

<app> <name>llr321</name> <fraction_done_exact/> <max_concurrent>8</max_concurrent> </app> <app_version> <app_name>llr321</app_name> <cmdline>-t 1</cmdline> <avg_ncpus>1</avg_ncpus> </app_version>


And, it now is able to run max (i.e. 8) tasks of LLR321 on this hosts in parallel each using one thread instead of running one task on 8 threads.

Tuna
4) Message boards : Problems and Help : My BOINC went haywire! (Message 134548)
Posted 393 days ago by Profile Tuna Ertemalp


Perhaps you have this option activated?


Yes, I have that option checked, and on purpose. I hadn't thought that 321 Sieve would be an empty task pipe. SInce then, seems it is delivering tasks.

Nevertheless, LLR321 claiming that it is MT, and not using all the threads I give to it is probably still a problem. In the meantime, I will change my app_config to NOT give LLR321 more than 1 threads so that multiple LLR321 jobs can execute in parallel at the expense of each job executing as fast as it could have.

Tuna
5) Message boards : Problems and Help : My BOINC went haywire! (Message 134370)
Posted 398 days ago by Profile Tuna Ertemalp
I think I am seeing something similar, and I didn't participate in any challenges. Noticed on at least one host that it is getting 321LLR only even though the machine is set to receive only 321-Sieve and DYFL tasks. There is at least another machine like that. Worth noting that I have not changed anything on my BOINC hosts for months. In other words, I didn't change anything either on this site or on my hosts regarding what tasks get downloaded, what venue the machines belong to, or any of the app_config files. Yet, these 321LLR tasks are the only non-DYFL tasks coming down, and one of those tasks is making BOINC stop using any other CPU cores since my app_config has the following on this 4-core HT host that hadn't changed since June 2018, yet the overall CPU usage per task is around 15-25%, not 100%, meaning -t 6 isn't doing anything and max_current is keeping other cores chillin':

<app>
<name>llr321</name>
<fraction_done_exact/>
<max_concurrent>1</max_concurrent>
</app>
<app_version>
<app_name>llr321</app_name>
<cmdline>-t 6</cmdline>
<avg_ncpus>6</avg_ncpus>
</app_version>

This is hostID 488599. But I am seeing the same on 510142, 927928 etc. I didn't check the others.

So, something is amiss?

Tuna
6) Message boards : Number crunching : LLR 3.8.20 Going Live! (Message 108021)
Posted 1292 days ago by Profile Tuna Ertemalp
Seems I forgot to include 321 in my previous app_config.xml. So, this should also be included:

<app> <name>llr321</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llr321</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version>


Tuna

Combining few posts in another thread:

http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107580
http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107589
http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107592
http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107594
http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107595

I created the below that I am now using on all of my 4/8/12/16 thread hosts.

Note that you cannot use it as-is; some minimal editing is required. You will need to:

1) Totally exit BOINCMgr (not just minimize!)
2) Create a {BOINCDataFolder}\projects\www.primegrid.com\app_config.xml, copy/paste this into it, and then ReplaceAll "NumProcs" with whatever value you want it to be. If you decide to give different NumProcs to different apps, then make sure to realize that each app occurs twice in this file, and to make sure each pair is consistent with each other.
3) Restart BOINCMgr

Also, when you restart BOINCMgr, you might get bunch of errors/warnings about stuff like "Your app_config.xml file refers to an unknown application 'llrSOB'. Known applications: 'pps_sr2sieve', 'ap26', 'llrPSP', 'llrTPS', 'trp_sr2sieve', 'llrPPSE', 'llrWOO', 'llrCUL', 'genefer', 'llrPPS', 'llrSR5', 'llrMEGA', 'gcw_sieve', 'llrGCW'". These are ignorable. All that means is that your host is probably set to receive tasks only from a subset of projects, which you probably manually handpicked using http://www.primegrid.com/prefs.php?subset=project, therefore to that host, not all projects are known. BOINCMgr just reports this, ignores it, and uses whatever is useful to it.

I am sure I will stand corrected if I made a mistake. There are a few warnings about doing this both in this thread as well as Michael's http://www.primegrid.com/forum_thread.php?id=7415&nowrap=true#107595. So, be careful. For now, I max'd each out on all my hosts. I'll see over time if that was deadly...

Tuna

<app_config> <app> <name>llrGCW</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrGCW</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrWOO</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrWOO</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrCUL</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrCUL</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrPSP</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrPSP</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrSOB</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrSOB</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrTRP</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrTRP</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrSR5</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrSR5</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> <app> <name>llrESP</name> <fraction_done_exact/> <max_concurrent>1</max_concurrent> </app> <app_version> <app_name>llrESP</app_name> <cmdline>-t NumProcs</cmdline> <avg_ncpus>NumProcs</avg_ncpus> </app_version> </app_config>

7) Message boards : Number crunching : LLR 3.8.20 Going Live! (Message 108020)
Posted 1292 days ago by Profile Tuna Ertemalp
You will need to:

1) Totally exit BOINCMgr (not just minimize!)
2) Create a {BOINCDataFolder}\projects\www.primegrid.com\app_config.xml, copy/paste this into it, and then ReplaceAll "NumProcs" with whatever value you want it to be. If you decide to give different NumProcs to different apps, then make sure to realize that each app occurs twice in this file, and to make sure each pair is consistent with each other.
3) Restart BOINCMgr


BOINC restart is not needed when you create or modify app_config.xml. It is enough to do this:
- create/update app_config.xml: file;
- disable option "Leave non-GPU tasks in memory while suspended" (Options/Computing preferences/Disk and memory tab);
- suspend all PrimeGrid tasks (or easier suspend CPU usage);
- reload config files (Options/Read config files);
- resume tasks.

Note: after doing this GUI still will display previous CPU count for apps. This is display bug only, number of running tasks and their command line will be adjusted properly.


I guess I found...

- File/Exit
- Create app_config.xml
- DblClick on BOINCManager

... to be easier. ;)

Tuna
8) Message boards : Proth Prime Search : PPS Sieve slower on 1080Ti vs TitanX?? (Message 107806)
Posted 1300 days ago by Profile Tuna Ertemalp
Valid: http://www.primegrid.com/results.php?hostid=510142&offset=0&show_names=0&state=4&appid=9
Pending: http://www.primegrid.com/results.php?hostid=510142&offset=0&show_names=0&state=2&appid=9


Yesterday (May 9 evening, PST), in this host, I replaced the two older Maxwell Titan X (~1550Mhz) with two 1080 Ti (~1950Mhz), and pretty much all CUDA apps (GFNs, especially) got much faster. However, all the PPS Sieve tasks seem to have gone slower, from ~350s+ to ~450s+, i.e. ~30%.

Thoughts? Just an anomaly I am seeing with the small sample size I have, or is this a systemic issue?

Tuna
9) Message boards : Generalized Fermat Prime Search : GFN22: 1080Ti vs. TitanZ (Message 107805)
Posted 1300 days ago by Profile Tuna Ertemalp
Given that I will be losing 4 GPUs in 2 Zs, but instead get 12 Tis overall as I replace all my GPUs with Tis, essentially making all of my GPUs GFN22-worthy, plus the power consumption & heat generation of Z being larger than X, plus the overall betterness of the TitanXs across the greater BOINC landscape of projects, seems moving few TitanXs down to older boxes as space/thermal/power envelopes allow would still make more sense.

Tuna
10) Message boards : Generalized Fermat Prime Search : GFN22: 1080Ti vs. TitanZ (Message 107800)
Posted 1301 days ago by Profile Tuna Ertemalp
GTX TITAN Z: ~60 hours (OCL: Complex numbers / 64-bit float)
GTX 1080 Ti: ~33 hours (OCL4: Number-Theoretic Transform / 31-bit integer)

(see http://www.primegrid.com/results.php?hostid=841181&offset=0&show_names=0&state=0&appid=17)


This settles it, then... No point in keeping the Zs. Now I can turn on GFN22 on all my 1080Ti cards... Eventually I'll have 12. ;) Plus move 2-4 Xs to older machines. Sell the rest.

Tuna


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.01, 0.02, 0.05
Generated 1 Dec 2020 | 22:20:03 UTC