Store at least n days of work ( for a 32 CPU device) : optimization ?

Questions and Answers : Preferences : Store at least n days of work ( for a 32 CPU device) : optimization ?
Message board moderation

To post messages, you must log in.

AuthorMessage
I3APR

Send message
Joined: 23 Apr 16
Posts: 99
Credit: 70,717,488
RAC: 0
Italy
Message 1784206 - Posted: 2 May 2016, 13:02:55 UTC

Hi there, I've been trying to fine tune this parameter, but I can't seem to get anywhere :

I can connect once per day, sometimes once every 3 days so I need to feed the "beast" with enough units to crunch inbetween.

The workload is 32 units at the same time, and it takes about 3 hours to complete each one, so, in order to stay isolated for three days, I need at least 240/260 units stored.

I cannot seems to get this far : even with network always active ( which eventually will be not), it is "n files uploaded" "- "n files downloaded" : ratio is always 1 : 1

Boinc is using "LOCAL" preferences, and I instructed it to store at least 5 days with 2 more additional.
Restarted Boinc, restarting the crunching, after setting the above parameter, with no luck.

What am I doing wrong ?

Thank you for your help !!

G.
ID: 1784206 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1784211 - Posted: 2 May 2016, 13:33:31 UTC - in response to Message 1784206.  

Seti has a 100 tasks for CPU, 100 tasks per GPU policy in place. Which means that you can for a 32 core CPU at max get 100 tasks in cache.

This is done because the results database has grown so big (several terabytes) that if they'd allow more tasks out there, it would just crash and take the project with it. It's done that once or twice before.

So the option would be here to add another project, perhaps Seti Beta?
ID: 1784211 · Report as offensive
I3APR

Send message
Joined: 23 Apr 16
Posts: 99
Credit: 70,717,488
RAC: 0
Italy
Message 1784223 - Posted: 2 May 2016, 14:11:34 UTC - in response to Message 1784211.  
Last modified: 2 May 2016, 14:14:00 UTC

Thank you for the quick reply...to say the truth, the system is a 8 physical CPU, 32 virtual....it should get 800 shouldn't it ?
ID: 1784223 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1784224 - Posted: 2 May 2016, 14:23:06 UTC - in response to Message 1784223.  

Starting with Seti Beta, it's a project on its own, but it runs and reruns work to test the science applications for Seti@Home. The results of the work done there are not added to the results database of Seti.
Credit for it is stored separate from the main project.
And if you like to use optimized applications, they'd rather you don't do that on their project as else it doesn't help towards developing the applications. You can also expect work to have a higher fault tolerance here than at Seti@Home.

But other than that, they're developed by the same people, housed in the same building, stored on the same servers. The address to add in BOINC Manager to add the project, and for its front page site is https://setiweb.ssl.berkeley.edu/beta/. When you add it with the same email address you use for here, you'll get the "volunteer tester" title under your name, at both projects.

But no, if you prefer to focus on a single project, if only to see how well your CPU does things, perhaps another project is better. But watch out, not all projects necessarily run on Windows Server, and not all are geared towards running with so many cores.

I'm a moderator on the BOINC forums and there I am helping a person who is running e.g. The Lattice Project and ATLAS@Home, and these both apparently are known to ask for a lot of memory per task, up to 8GB. Just as a warning that even with a 32 core CPU and 32GB RAM you could get into trouble rather fast then.

A good stroll through the projects list may be what's needed. :)
ID: 1784224 · Report as offensive
Profile Jord
Volunteer tester
Avatar

Send message
Joined: 9 Jun 99
Posts: 15184
Credit: 4,362,181
RAC: 3
Netherlands
Message 1784226 - Posted: 2 May 2016, 14:27:29 UTC - in response to Message 1784223.  
Last modified: 2 May 2016, 14:27:45 UTC

Thank you for the quick reply...to say the truth, the system is a 8 physical CPU, 32 virtual....it should get 800 shouldn't it ?

BOINC doesn't count the physical CPUs, only the logical CPUs as that's the only API they have found that's of use.

David Anderson wrote:
The BOINC client uses GetActiveProcessorCount() to get the # of CPUs:
https://msdn.microsoft.com/en-us/library/windows/desktop/dd405485%28v=vs.85%29.aspx
This returns the # of "logical processors" (i.e. a HT CPU counts as 2).
I can't find an API for getting the # of physical processors.

ID: 1784226 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1784248 - Posted: 2 May 2016, 16:37:35 UTC

To ensure you get maximum loading at each server connection it is best to have a very small "additional days", 0.1 will give you a fair chance of getting up to the 100 for your CPU, plus 100 per GPU on the system.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1784248 · Report as offensive
I3APR

Send message
Joined: 23 Apr 16
Posts: 99
Credit: 70,717,488
RAC: 0
Italy
Message 1784716 - Posted: 4 May 2016, 13:33:43 UTC - in response to Message 1784224.  


But no, if you prefer to focus on a single project, if only to see how well your CPU does things, perhaps another project is better. But watch out, not all projects necessarily run on Windows Server, and not all are geared towards running with so many cores.
....
A good stroll through the projects list may be what's needed. :)



Thank you Jord, very exhaustive as a reply...it's a matter of "affection" : I returned to SETI after 13 years ( was n. 1 in my country back then :-D ), and I believe I will stick to it.
Of course it is nobody's fault but mine, having such narrow connection schedules during the day, although I wished the client had more granularity in programming the computing/connection schedule.
Having 32 CPU per device is rather a disadvantage it seems, but I'll try to harvest as much credit as I can...( I reckon I'm loosing about 20/30 % of computing time ).
Thank you again for the kind reply !

G.
ID: 1784716 · Report as offensive
I3APR

Send message
Joined: 23 Apr 16
Posts: 99
Credit: 70,717,488
RAC: 0
Italy
Message 1784717 - Posted: 4 May 2016, 13:35:48 UTC - in response to Message 1784248.  

To ensure you get maximum loading at each server connection it is best to have a very small "additional days", 0.1 will give you a fair chance of getting up to the 100 for your CPU, plus 100 per GPU on the system.


Did configure as you suggest, thank you ;-)

G
ID: 1784717 · Report as offensive

Questions and Answers : Preferences : Store at least n days of work ( for a 32 CPU device) : optimization ?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.