climateprediction.net home page
HADCM3N propensity to crash.

HADCM3N propensity to crash.

Message boards : Number crunching : HADCM3N propensity to crash.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Dave Jackson
Volunteer moderator

Send message
Joined: 15 May 09
Posts: 4531
Credit: 18,708,023
RAC: 16,577
Message 45484 - Posted: 19 Jan 2013, 9:46:58 UTC

I have two of these tasks running on this box, one at 74% and one at 84%. This computer is hibernated every night without suspending the tasks. They have also survived one power outage. Does this indicate that these units are no longer quite as susceptible to crashing as they were or does it mean that the linux hibernate has improved? The dual core atom also running linux is only hibernated once a week and as it runs so slowly it is a bit early to say how it is doing.
ID: 45484 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2184
Credit: 64,822,615
RAC: 5,275
Message 45485 - Posted: 19 Jan 2013, 16:43:43 UTC - in response to Message 45484.  

I don't think there's been any change to hadcm3n, so you are either 1 very lucky, or 2 linux hibernate has got better.

I'm guessing 1, but who knows.
ID: 45485 · Report as offensive     Reply Quote
Profile Dave Jackson
Volunteer moderator

Send message
Joined: 15 May 09
Posts: 4531
Credit: 18,708,023
RAC: 16,577
Message 45486 - Posted: 20 Jan 2013, 10:01:08 UTC

I have just thought of another couple of factors and have also looked at a couple of sites talking about the hibernate function.

Not sure but these may be the first coupled ocean tasks since I upgraded to BOINC 7.X.

I have also increased memory to 4GB from 2.

One of the sites I looked at suggested that there had been a bug in the kernel memory allocation, another suggested that older versions of hibernate had problems above 4GB of ram though as that is the maximum my board can take it doesn't affect me.

Less than 100 hours computing time to see if first task will complete now. Second task is about 10% behind the first.
ID: 45486 · Report as offensive     Reply Quote

Message boards : Number crunching : HADCM3N propensity to crash.

©2024 cpdn.org