Another credit reduction?
31 Aug 2006 18:32:32 UTC
Topic 191763
(moderation:
On my 4 hosts all newer WUs show a shorter time for calculation, even they are out of the same series.
A teammate already got a result with less credits.
Another credit reduction?
)
Good spot. I have a reasonably fast P4 currently working on short WUs (l1_0188.5_...), and there's a similar drop in estimated times with WUs downloaded after about 17:00 UTC yesterday (+/- 30 mins).
That's reading estimated times from my cache, but one of my quorum partners (717234) has started getting reduced credit.
Bernd/Bruce, I appreciate what you're trying to do with the fair credit across projects, but I'm not sure it's a good idea to slip these changes in without comment - it can make the natives restless...
yes... the first credit drop
)
yes... the first credit drop was ok and needed, the second one was already a bit... hmm.. and this one is really too much i think....
RE: yes... the first credit
)
Yes way to much but they don't care anyway....
Anyway did someone ever tried a statistics regarding the output (RAC) if a computer only runs long wus compared to if it would run only small ones ?
RE: Bernd/Bruce, I
)
This failure to communicate is my fault.
A couple of days ago, after studying the average credit/hour of E@H hosts and comparing with other projects, I came to the conclusion that we were granting about 8% too much credit, so I reduced the granted credit by this amount.
Bruce
Director, Einstein@Home
RE: RE: Bernd/Bruce, I
)
Bruce,
I just don't understand how you can figure out an average credit/hour among hosts.
Here are my hosts (all with XP, 2400+ with Linux) :
Opteron 170 : 22.7 per hour per cpu => 45.55 per hour for this host
P4C 3.2 (HT ON) : 8.83 per hour per cpu => 17.66 per hour for this host
XP 2400+ : 18.31 per hour per cpu => 18.31 per hour for this host
P4B 2.4 : 13.26 per hour per cpu => 13.26 per hour for this host
Pentium-m 1.6 (Banias) : 13.95 per hour per cpu => 13.95 per hour for this host
I guess the difference between 2 of my hosts could be call a gap. A huge one!
45.55 vs 13.26 : a factor of almost 3.5
Not easy to figure out an "average" with such figures...
From what I get for my hosts, my server should remain on this project.
P4C and XP2400+ may have a look elsewhere to see if they can get better
P4B and P-M, well, they SHOULD get their job from another project.
With these kind of "drastic" average credit per hour, I guess that we may have to improve Boinc. Something like detecting the cpu and giving an automatic advice to the user : "your host is well suited for a, b and c projects. It is OK for d and e. It is NOT recommanded for any other project".
In my case, I passed from Seti to Einstein after the "events" concerning optimized clients. I guess my intel hosts may go back there and get the credit they diserve. As many people (I guess), the more credits my hosts get, the more I think they are usefull for the project. I let you conclude what I think about my intel boxes on E@H...
Boxes are different, as projects are. Some grant more credit on some projects, some grant less. So trying to get an "average credit/hour/host" is vain.
An average could also be done by dividing the credit/hour of the entire E@H project by the number of active hosts. It would be fair (sarcastic)
Agree with black hole. If
)
Agree with black hole.
If E@H will continue like this then credits granted will have to be reduced constantly if new powerful hosts join in here at a higher rate than in other projects. In other words, every time someone sets up a new big fat cruncher we should get less credit. Comparing with other projects is probably good (and totally unnecessary), but how come i see far less Pentium III's here than in for example Rosetta? I mean, just look at the hosts of for example Bruce himself, so many dualcore Opterons is by itself enough to give E@H higher average crunching power per host compared to most other projects. It sure seems like this last credit reduction was too much, i think around 8% too much?
Increase of total crucnching power at Einstein = decrease of credits granted.
Team Philippines
That's a fair point you two
)
That's a fair point you two guys make. I never thought about it before but it may well be possible that Einstein simply has (on average) more powerful computers than most other projects and therefore it would be deserved if they got a bit more credit... no idea if there is a reliable way to figure this out but it may be worth considering.
From BoincStats.com E@H
)
From BoincStats.com
E@H :
70919.2 GigaFLOPS
73904 active hosts
=> 0.956 GigaFLOPS / Host
S@H :
205273.1 GigaFLOPS
271500 active hosts
=> 0.756 GigaFLOPS / host
Guess what ?
Active hosts on E@H are 21% more powerfull than on S@H.
S@H is just an example. You can try with the rest of the projects.
I repeat : credit is not fair on E@H
Bruce wants (or has to) make it fair regarding other projects, but hosts are not the same !
As a consequence, attach you P3's to E@H and your top boxes to other projects if you want to grant credit. Stupid !
RE: yes... the first credit
)
Hi Cruncher
I did it.
the difference is about 10% : about 1-2 credit / hour on a box running about 20 credits / hour
RE: RE: yes... the first
)
So you're saying that smaller Wus give us back what the project has stolen the last time ? 10 % ?
If that's true i know how to deal with it.