Long runtime ECM WU with short deadlines

Fehler und Wünsche zum Projekt yoyo@home
Bugs and wishes for the project yoyo@home
Nachricht
Autor
Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: Long runtime ECM WU with short deadlines

#13 Ungelesener Beitrag von yoyo » 28.12.2016 17:26

I'm not fully clear how boinc works regarding memory.
I assume boinc checks that first wu requires 10 GB of ram and at least 10 GB are free. So this wu is started.
But it needs (currently) only 2.5 GB.
So boinc checks the next wu and so on.
So it could be that boinc starts all 4.

I would assume if some WU grows up to 10 GB and there isn't enough memory for the other WU, than boinc stops the wu with "waiting for memory".
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Dunckx
PDA-Benutzer
PDA-Benutzer
Beiträge: 45
Registriert: 12.11.2014 09:26

Re: Long runtime ECM WU with short deadlines

#14 Ungelesener Beitrag von Dunckx » 28.12.2016 20:33

This is getting interesting!

I just had five P2 WU running on 32GB of ram.

I was using Firefox and a Windows dialogue box came up suggesting I close ecm because there was too little memory. This prompted me to look at the BOINC window and indeed one WU was "waiting for memory". Then the tab crashed in Firefox and one of the other P2 terminated with "computation error". Then another P2 WU terminated with the same error. I was just fast enough to grab a screen shot of this WU with the computation error as another P2 WU started up to replace it. So there's now only three P2 WU running and enough ram to go round.

Somehow I reckon BOINC needs to be a bit smarter in choosing which WU can run simultaneously. Five P2 at once in 32GB is just not going to be happy. Three seems OK though, it's the queueing of more than three in a row which is problem #1. If only there was a way to pad them out with other WU in between that might work, but I suspect that is an issue for the BOINC team.

Dunckx
Dateianhänge
P2 WU crash.jpg
Too many P2 WU at once
P2 WU crash.jpg (72.22 KiB) 6657 mal betrachtet

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: Long runtime ECM WU with short deadlines

#15 Ungelesener Beitrag von yoyo » 28.12.2016 23:14

In Boinc you can limit the usage of memory for Boinc. You should set this to 75%. So there is always RAM available for other processes.
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Benutzeravatar
helcoin
PDA-Benutzer
PDA-Benutzer
Beiträge: 46
Registriert: 10.01.2016 22:46
Wohnort: Stuttgart

Re: Long runtime ECM WU with short deadlines

#16 Ungelesener Beitrag von helcoin » 29.12.2016 13:14

Hi,
This job will take about 180 hours to run. A very long runtime, is this ok?
yoyo_175std.PNG
Estimate runtime will be about 180 hours?
yoyo_175std.PNG (14.35 KiB) 6629 mal betrachtet
info.txt:
Name: C305_142_139xyT3e25
Length: 305
B1: 25000000000
Curve: 500

Project:
Duration correction factor is 5,1631
Bild
Bild

respawner
Vereinsmitglied
Vereinsmitglied
Beiträge: 554
Registriert: 10.12.2007 19:42

Re: Long runtime ECM WU with short deadlines

#17 Ungelesener Beitrag von respawner » 29.12.2016 13:40

My P2-WUs needed about 12.5 hours. The prediction was 6 days.
BildBild

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: Long runtime ECM WU with short deadlines

#18 Ungelesener Beitrag von yoyo » 29.12.2016 21:20

helcoin hat geschrieben:Hi,
This job will take about 180 hours to run. A very long runtime, is this ok?

info.txt:
Name: C305_142_139xyT3e25
Length: 305
B1: 25000000000
Curve: 500

Project:
Duration correction factor is 5,1631
I do not have any result yet of this series, but it is predicted with 80h on an 5 year old Linux Computer.
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Dunckx
PDA-Benutzer
PDA-Benutzer
Beiträge: 45
Registriert: 12.11.2014 09:26

Re: Long runtime ECM WU with short deadlines

#19 Ungelesener Beitrag von Dunckx » 29.12.2016 23:08

It's beginning to look like it was the memory available when the computer was idle which was the issue. I had a look at the BOINC computing preferences and the maximum memory during pc use was already set at 75%. However, when the computer is idle, the memory availability to BOINC was 100%. I have now reduced this to 90% and no more crashes.

It begs the question, how exactly is BOINC determining that the computer is "idle". When those two P2 WU crashed and the Firefox tab went with them, I was watching something on YouTube. No mouse movements, no keyboard input. Under the then computing preferences, BOINC may well have had up to 100% of the ram. It's a pity I didn't have Task Manager running at the time.

OK, file under "interesting, but not necessarily useful."

Dunckx

Woof
Idle-Sammler
Idle-Sammler
Beiträge: 3
Registriert: 31.12.2016 09:17

Re: Long runtime ECM WU with short deadlines

#20 Ungelesener Beitrag von Woof » 31.12.2016 09:36

yoyo hat geschrieben:I do not have any result yet of this series, but it is predicted with 80h on an 5 year old Linux Computer.
I am actually running several of this series. Between a Xeon E5-2620 and some X5570s I am seeing estimates of between 41-53 days of compute time.

I have around 3 or so of these that have been running now for ~52 hours or so between 5-7% done. Most of these are due on the second of January.

I'll keep them going and try and keep track of what is happening but I don't want to end up losing ~100hrs of compute time

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: Long runtime ECM WU with short deadlines

#21 Ungelesener Beitrag von yoyo » 31.12.2016 21:25

I have now some results. P1 of T3e25 runs between 30 and 70 hours.
P2 between 14 and 40 hours.
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Dunckx
PDA-Benutzer
PDA-Benutzer
Beiträge: 45
Registriert: 12.11.2014 09:26

Re: Long runtime ECM WU with short deadlines

#22 Ungelesener Beitrag von Dunckx » 31.12.2016 22:46

OK, I currently have eight of these running, two days elapsed, five more estimated and only two or three days left prior to the deadline. It will be interesting to see how long these take in total and whether they meet the deadline requirement.

Ein gutes neues Jahr!

Dunckx
Dateianhänge
8x as T3e25 WU.jpg
Eight as T3e25 WU
8x as T3e25 WU.jpg (82.76 KiB) 6560 mal betrachtet

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: Long runtime ECM WU with short deadlines

#23 Ungelesener Beitrag von yoyo » 31.12.2016 22:50

They will finish next year ;)
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Benutzeravatar
Beyond
Prozessor-Polier
Prozessor-Polier
Beiträge: 111
Registriert: 02.02.2008 01:48
Wohnort: Rum River watershed, MN, USA

Re: Long runtime ECM WU with short deadlines

#24 Ungelesener Beitrag von Beyond » 31.12.2016 23:29

yoyo hat geschrieben:They will finish next year ;)
Actually I bet that they will all fail or have already failed and are stuck unreported in BOINC. People should check their machines to make sure they're REALLY still running. See my post here:

viewtopic.php?f=57&t=16511&p=167498#p167498

Antworten

Zurück zu „Fehler, Wünsche / Bugs, Wishes“