Credit equality soon?
26 Mar 2012 17:34:50 UTC
Topic 196254
(moderation:
In message 113483, Bernd stated (with respect to the GRP/GW credit discrepancy:
Quote:
As BOINC's "Credit New" system has not shown to work reliably, E@H still assigns credit (and flops estimation) manually on the server side, which requires some fiddling and tuning, especially at the beginning of using a new application.
Also too, we are still tuning the validator.
Please bear with us, we're doing our best.
And as Bikeman stated in message 113512:
Quote:
Yes, the crediting of the LAT WUs needs some more fixing.
Both of these quotes are from early August 2011, and we are still waiting. I participate in E@H to, 1)Contribute to the science, and 2)To put my computer to good use. So credits are not a high priority for me, but credits are the currency for keeping track of one's contribution.
I went back one year to research this topic. It is now nearly 6 months since the above messages were posted. Has there been any 'movement' to correct the discrepancy? Why is it taking so long, and when do you think 'fairness' will be achieved?
Steve
Edit: As an example of the discrepancy, the average of five recent GRP WUs is 21,446s (CPU time) for 158 credits, whereas the average of ten recent GW WUs is 14,070s for 251 credits. So GRP WUs take 1.5 times as long, but are only 'worth' 0.63 times as many credits. This is equivalent to 136 seconds/credit for GRP, and 56 seconds/credit for GW. Therefore, the ratio of GRP to GW seconds/credit is 2.38.
"Remember, nothing that's good works by itself, just to please you. You have to make the damn thing work." Thomas A. Edison
Credit equality soon?
)
If somebody care, Im not crunching LAT WUs due to this discrepancy.
Even when I dont care much on credits, I still want them as they are the only reward I get for contributing (I know that credit or not, there is science beneffit but thats not something that I get for myself.)
RE: As an example of the
)
Hi!
Please check again:
GRP tasks are awarded 337 credits (granted credit! Ignore the "claimed credit")
GW tasks are worth 251 credits as you wrote.
So the ratio for your PC (it varies with CPU model) is more like
[pre]
21446 14070
----- / ------ ~ 1.14
337 251
[/pre]
HBE
RE: //... So the ratio for
)
Its still 14% more credits awarded by the GW over the GRP... I have one i7-860, one i7-2600 and one Core2 E8400 and while the crunching times are different, the ratio is more or less the same in all of them...
Anyway, this is not meant to be a complain and I think that may be this is on pourpouse to get more people interested on crunching GWs than GRPs...
If thats the goal they got me! :D
RE: RE: //... So the
)
Not all CPU architectures are the same; as a result identical payouts on one platform will just magnify the divergence on others. IIRC S6 does significantly better on Intel than on AMD (I know S5 did). I don't have any operational AMD boxes, but I think they'd show BRP4-CPU and FGRP as outscoring S6. At this point I think PPC (old macs) is too small a share to be worth considering for balancing purposes. At times in the past there were significant variations between OSes too; but I think they mostly went way after the project was able to get all platforms building under GCC.
FWIW, this is what I get when
)
FWIW, this is what I get when looking at the valid results on my host (i7).
[pre]
Application Average cr/s Tasks
Gamma-ray pulsar search #1 v0.23 0.011642362 22
Gravitational Wave S6 GC search v1.01 (SSE2) 0.013866201 10
Gravitational Wave S6 LineVeto search v1.10 (SSE2) 0.013585561 31
[/pre]
Comparing the GW results against the pulsar search results, that makes GW crunching almost 17% more profitable.
RE: Not all CPU
)
Thats a good point ...
Anyway, BRP4-CPU gives an awfully low credit in my Intel hosts compared to GW.
RE: Thats a good point ...
)
I thought BRP tasks are send to GPUs only?!
Michael
Team Linux Users Everywhere
RE: RE: Thats a good
)
Only if you deselect the "Run CPU versions of applications for which GPU versions are available" preference.
RE: RE: RE: Thats a
)
AFAIK, there was a time in wich BRPs were only sent to GPUs despite the preferences, but it was a temporary workaround to deal with some issue that I cant remember... (I think it was something related to the inefficient way Apache servers handled the resources needed to transfer hughe files through http... but may be Im wrong and confusing things...)