An idea to resolve checkpointing problem...

Everything about the project RNA World
Nachricht
Autor
szopler
Mikrocruncher
Mikrocruncher
Beiträge: 27
Registriert: 31.01.2013 16:32

An idea to resolve checkpointing problem...

#1 Ungelesener Beitrag von szopler » 31.01.2013 16:41

I've tried to use virtualbox virtual machine with Debian inside to crunch RNA and also its snapshot mechanism to prevent lose of data when I turn of the computer.
It works fine so i think that you should modify your project to use virtual machine with automatic snapshots instead of no checkpointing 10 (and more) day LONG workunits.

PS.
In project Climate@Home it works fine :)

ChristianB
Vereinsvorstand
Vereinsvorstand
Beiträge: 1914
Registriert: 23.02.2010 22:12

Re: An idea to resolve checkpointing problem...

#2 Ungelesener Beitrag von ChristianB » 31.01.2013 18:47

Thanks for this tip. We already discussed the new virtualboxwrapper feature of BOINC but don't have the manpower to test and implement into RNA world at the moment. As it is not easy for volunteers to set up a Virtual Machine on there own and there are no easy works-for-all instructions this approach is for the more experienced volunteers only.

Benutzeravatar
Michael H.W. Weber
Vereinsvorstand
Vereinsvorstand
Beiträge: 20230
Registriert: 07.01.2002 01:00
Wohnort: Marpurk
Kontaktdaten:

Re: An idea to resolve checkpointing problem...

#3 Ungelesener Beitrag von Michael H.W. Weber » 02.02.2013 19:59

In fact, I am planning to write up a brief instruction on how to install a VM manually with RNA World.
Many people, however, do *NOT* like to use VMs especially in DC because it is less effective computation-wise (7-15% reduced throughput would be the approximate price for having checkpointing with such an approach).

Michael.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

http://signature.statseb.fr I: Kaputte Seite A
http://signature.statseb.fr II: Kaputte Seite B

Bild Bild Bild

jon b.
Mikrocruncher
Mikrocruncher
Beiträge: 17
Registriert: 30.01.2013 06:20

Re: An idea to resolve checkpointing problem...

#4 Ungelesener Beitrag von jon b. » 05.03.2013 01:31

Does version 1.1 of Infernal support checkpointing?

jon b.
Mikrocruncher
Mikrocruncher
Beiträge: 17
Registriert: 30.01.2013 06:20

Re: An idea to resolve checkpointing problem...

#5 Ungelesener Beitrag von jon b. » 06.03.2013 02:00

Michael H.W. Weber hat geschrieben:In fact, I am planning to write up a brief instruction on how to install a VM manually with RNA World.
Many people, however, do *NOT* like to use VMs especially in DC because it is less effective computation-wise (7-15% reduced throughput would be the approximate price for having checkpointing with such an approach).
7-15% reduced throughput is better than 100% computation loss due to reboot, power failure, etc.

Benutzeravatar
Michael H.W. Weber
Vereinsvorstand
Vereinsvorstand
Beiträge: 20230
Registriert: 07.01.2002 01:00
Wohnort: Marpurk
Kontaktdaten:

Re: An idea to resolve checkpointing problem...

#6 Ungelesener Beitrag von Michael H.W. Weber » 06.03.2013 12:09

jon b. hat geschrieben:Does version 1.1 of Infernal support checkpointing?
No. But 100-fold performance increase.
jon b. hat geschrieben:
Michael H.W. Weber hat geschrieben:In fact, I am planning to write up a brief instruction on how to install a VM manually with RNA World.
Many people, however, do *NOT* like to use VMs especially in DC because it is less effective computation-wise (7-15% reduced throughput would be the approximate price for having checkpointing with such an approach).
7-15% reduced throughput is better than 100% computation loss due to reboot, power failure, etc.
Of course, but still there will be people not willing to use it. Which is OK for me.

Michael.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

http://signature.statseb.fr I: Kaputte Seite A
http://signature.statseb.fr II: Kaputte Seite B

Bild Bild Bild

jon b.
Mikrocruncher
Mikrocruncher
Beiträge: 17
Registriert: 30.01.2013 06:20

Re: An idea to resolve checkpointing problem...

#7 Ungelesener Beitrag von jon b. » 06.03.2013 16:12

Michael H.W. Weber hat geschrieben:
jon b. hat geschrieben:Does version 1.1 of Infernal support checkpointing?
No. But 100-fold performance increase.
Michael.
Will this version be implemented?

Benutzeravatar
Michael H.W. Weber
Vereinsvorstand
Vereinsvorstand
Beiträge: 20230
Registriert: 07.01.2002 01:00
Wohnort: Marpurk
Kontaktdaten:

Re: An idea to resolve checkpointing problem...

#8 Ungelesener Beitrag von Michael H.W. Weber » 09.03.2013 11:35

Yes.

Michael.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

http://signature.statseb.fr I: Kaputte Seite A
http://signature.statseb.fr II: Kaputte Seite B

Bild Bild Bild

jon b.
Mikrocruncher
Mikrocruncher
Beiträge: 17
Registriert: 30.01.2013 06:20

Re: An idea to resolve checkpointing problem...

#9 Ungelesener Beitrag von jon b. » 25.03.2013 00:05

When version 1.1 is implemented, it would be nice if running the applications in multithreaded mode could be enabled/disabled, and the number of cores used set in the project settings.

Benutzeravatar
Michael H.W. Weber
Vereinsvorstand
Vereinsvorstand
Beiträge: 20230
Registriert: 07.01.2002 01:00
Wohnort: Marpurk
Kontaktdaten:

Re: An idea to resolve checkpointing problem...

#10 Ungelesener Beitrag von Michael H.W. Weber » 25.03.2013 22:28

The 100-fold speed increase relates to single core computation, so a quad core machine could reach 400-fold. :wink:

Michael.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

http://signature.statseb.fr I: Kaputte Seite A
http://signature.statseb.fr II: Kaputte Seite B

Bild Bild Bild

Antworten

Zurück zu „RNA World Discussions (english)“