Change in Cobblestones ?

eOn code for long time scale dynamics

Moderator: moderators

chill
Posts: 96
Joined: Tue Jul 28, 2009 9:04 pm

Re: Change in Cobblestones ?

Post by chill »

I am still looking into the credit issue. I am having a lot of trouble understanding how this can happen as we haven't modified anything about the way credit is granted. I am writing an email today to the BOINC developers mailing list to inquire about what might be going wrong. It might be a coincidence, but these problems seemed to happen around the same time that many of our clients started to exceed the specified maximum floating point operations, which didn't make sense as our workunits were very short compared to the set maximum.

I am sorry for the lack of information, but we are trying to resolve the issue.
chill
Posts: 96
Joined: Tue Jul 28, 2009 9:04 pm

Re: Change in Cobblestones ?

Post by chill »

nullcoding wrote:Actually, I take that back (sorta). One of my XP boxes is now showing time estimates of over an hour per task and is running eOn as "high priority," meaning everything else gets shunted off the CPU. Tad annoying, that. I'll see how many credits I get for those, but geez, I hope it doesn't take an hour for a 5-credit task that runs in about 10 minutes on 64-bit OS X.

Which reminds me, I meant to ask if eOn is optimized for 64-bit OSs. If so, then my questions are answered. If not, then...?

I can understand this might be due to my use of rather dated hardware, but it runs other projects just fine (albeit in ~3x the time).
We are having some issues with older cpus running slower than I think they should be. Currently and the only simulation running is generating workunits that should be taking between about 2 - 10 minutes on my laptop (2.8 GHz intel core 2 duo, max osx, ~3Gigaflops/core).

In reply to your question about 64-bit OSs, we have 64 bit binaries for linux and osx. We have 32bit binaries for windows, but no 64bit binaries. This doesn't matter much at all as our application shouldn't speed up by being compiled as a 64bit binary. There are some cpu features on newer cpus that we might be able to take advantage of in order to speed up calculations on newer cpus, but it probably wouldn't make a big difference.

Could you provide the cpu info of your windows machine(s) that are running slowly? Or better yet the BOINC host id?

This morning I did a calculation to determine the fairness of the credit each result was getting. Credit should be proportional to the product of how many CPU seconds were used and the benchmarked floating point operations of the host computer. Thus the cpu_time*flops/credit should be a constant number for every workunit calculated if there is a linear relationship between the number of floating point operations a host does for a workunit. However, I see that some hosts get more credit than others for every floating point operation performed. I haven't found a good document describing how credit is calculated in BOINC so I think I will ask them in an email what is going on.
Saenger
Posts: 27
Joined: Thu Sep 02, 2010 4:23 pm
Contact:

Re: Change in Cobblestones ?

Post by Saenger »

If the correlation between crunch time and flops an any given machine is always the same, there is no correlation between flops and credits on my machine from app v2.03.
It was very much the a flat line for the v1.33 app.

BTW: Currently the results make good on the exorbitant granting from the last week, I'm down to 10-15 C/h with the last ones.
Saenger
Posts: 27
Joined: Thu Sep 02, 2010 4:23 pm
Contact:

Re: Change in Cobblestones ?

Post by Saenger »

Here's an overview of the credits on my machine since September 2010:

Code: Select all

Old server version	Runtime	CPU-Time	Claimed	Granted	C/h claim	C/h grant	C/h Diff
App name	number of WUs					CPU
eon client v5.04	156	40688,01	33394,65	255,35	255,35	27,53	27,53	0,00
eon client v5.08 	38	12109,15	9730,82	74,44	73,32	27,54	27,13	-0,41
eOn Client v1.00	8	1987,5	1405,86	10,82	10,82	27,71	27,71	0,00
New server version	Runtime	CPU-Time	Credits		C/h CPU	C/h clock	C/h Diff
App name	number of WUs							
eOn Client v1.00	621	87855,56	79666,29	946,18		42,76	38,77	-3,99
eOn Client v1.10 	1	575,37	543,34	5,54		36,71	34,66	-2,04
eOn Client v1.18	15	1227,05	1102,11	24,84		81,14	72,88	-8,26
eOn Client v1.19	64	22813,87	21599,61	256,32		42,72	40,45	-2,27
eOn Client v1.21	89	55190,13	52686,68	750,05		51,25	48,93	-2,32
eOn Client v1.22 	163	86671,3	82393,92	969,69		42,37	40,28	-2,09
eOn Client v2.03	126	31803,34	29979,8	1130,64		135,77	127,98	-7,78
My computer ist this one, hardware profile is here: http://www.sysprofile.de/id48630
Attachments
eOn WUs Saenger.ods
An OpenOffice-Calc file with data of my crunched WUs
(176.34 KiB) Downloaded 902 times
skipdashu
Posts: 3
Joined: Fri Nov 12, 2010 6:19 pm

Re: Change in Cobblestones ?

Post by skipdashu »

Chill,

Here's the pseudo code needed at the end of the credit calc routine. Please code & apply at your earliest convenience.

Code: Select all

If OS = Linux
    credit = credit * 2;

If User.Team = "Guru Mountain"
    credit = credit * 2
    If User.Zip = "78634"
        credit = credit * 2;
I'm sure it'll resolve all the credit questions / issues.

- da shu @ HeliOS,
"La carencia de recursos no debe de ser impedimento para que un niño tenga acceso a la tecnología."
Post Reply