CPU Time versus Credit

[DPC]Division_Brabant~Schaduwtje
[DPC]Division_B...
Joined: 29 Oct 08
Posts: 34
Credit: 5,526,816
RAC: 0
Topic 194830

The amount of credit awarded for completed S5GCE tasks versus the amount of CPU time spent on such a task seems to be a bit off.

Compared to the pulsar search work units, I sometimes see that the S5GCE take about twice as long, or even three times longer than the pulsar units. However the pulsar units even get a bit more credit than the S5GCE unit.

Is that a known fact and will something be done about it for us stats-hungry enthusiasts?

KSMarksPsych
KSMarksPsych
Moderator
Joined: 15 Oct 05
Posts: 2,702
Credit: 4,090,227
RAC: 0

CPU Time versus Credit

Quote:

The amount of credit awarded for completed S5GCE tasks versus the amount of CPU time spent on such a task seems to be a bit off.

Compared to the pulsar search work units, I sometimes see that the S5GCE take about twice as long, or even three times longer than the pulsar units. However the pulsar units even get a bit more credit than the S5GCE unit.

Is that a known fact and will something be done about it for us stats-hungry enthusiasts?

Yes

Kathryn :o)

Einstein@Home Moderator

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1,079
Credit: 341,280
RAC: 0

See Message 103079 - Posted

See Message 103079 - Posted 21 Mar 2010 1:37:52 UTC - in response to Message 103072:

Gary Roberts wrote:

What Bernd said in the sticky thread was this (emphasis is mine) -

Quote:
It would seem the GC tasks run a bit longer than we planned for. For the time being I raised the flops estimation and credit for newly generated tasks by a factor of 1.6.

The flops change will hopefully improve the crunch time estimate and the credit change will address your particular point of concern, but, unfortunately, both changes can only be seen in tasks generated on the server after the change was made. Existing tasks out in the field or cached on the server will still have the 'pre-change' values of both parameters and the change will only be seen once those tasks are cleared through the system.


Computer sind nicht alles im Leben. (Kleiner Scherz)

tullio
tullio
Joined: 22 Jan 05
Posts: 2,049
Credit: 41,465,911
RAC: 17,283

I am getting 19 credits/hour

I am getting 19 credits/hour on S5R6 and 14 credits/hour (round figures) on GCE. This on an Opteron 1210 running SuSE Linux 11.1 at 1.8 GHz. Just for the record, I am not complaining.
Tullio

Whiskymania
Whiskymania
Joined: 2 Sep 05
Posts: 8
Credit: 10,030,266
RAC: 0

I'm quite a bit upset finding

I'm quite a bit upset finding the same discussion like in August 2008, when
Credits were divided by - 2 (?)
Now it's again a division of arround 2 for these new Workunits GCS5E.

For me isn't anymore "just for the record" - I'm complaining now.
I couldn't understand the arguments back in August 2008 and so I cannot understand, why it is not possible, to get an addequate credit/WU for the new GCS5E workunits nowadays.
Based on the discussion in August 2008: Why shouldn't e@h be the project with the most credit/WU? Other projects (Milkyway, Collatz) support GPU and earn credits per WU that are a thousand times higher than e@h!

I'm not asking you why - I'm asking you to change.
I've started to delete the GCS5E WUs and will not stop, until a change has been made, cause I'm not willing anymore to just "take what e@h would like to give". I think it should change to be just a bit more faire, doesn't it?

I'm quite angry right at that moment and I think is't rather readable on the lines than between the lines.

Nitce on the side:
Cause we are a partie of at least three and I have to tell you that these words
are from just one member of our account an the others will, I'm quite shure
about that, not agree with me. So don't make "them" fool because of mine.

Markus/CH

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

RE: I'm quite a bit upset

Message 97415 in response to message 97414

Quote:

I'm quite a bit upset finding the same discussion like in August 2008, when
Credits were divided by - 2 (?)
Now it's again a division of arround 2 for these new Workunits GCS5E.

For me isn't anymore "just for the record" - I'm complaining now.
I couldn't understand the arguments back in August 2008 and so I cannot understand, why it is not possible, to get an addequate credit/WU for the new GCS5E workunits nowadays.
Based on the discussion in August 2008: Why shouldn't e@h be the project with the most credit/WU? Other projects (Milkyway, Collatz) support GPU and earn credits per WU that are a thousand times higher than e@h!

I'm not asking you why - I'm asking you to change.
I've started to delete the GCS5E WUs and will not stop, until a change has been made, cause I'm not willing anymore to just "take what e@h would like to give". I think it should change to be just a bit more faire, doesn't it?

I'm quite angry right at that moment and I think is't rather readable on the lines than between the lines.

Nitce on the side:
Cause we are a partie of at least three and I have to tell you that these words
are from just one member of our account an the others will, I'm quite shure
about that, not agree with me. So don't make "them" fool because of mine.

Markus/CH

Well, no offense intended, but then you should consider just suspending EAH entirely until we get started on the next GW search.

The reason is both ABP and GCE pay somewhat less generally speaking than GW does.

Also, comparing MW to EAH (or SAH for that matter) is inappropriate. One reason for that is that most hosts run a GPU app over there and that's not the case here at EAH. Also, MW has ALWAYS set a credit basis which is higher than EAH or SAH, even for the CPU version of the application.

Don't get me wrong here, I agree with you in principle. However, in my almost 5 years of experience with BOINC, EAH has made the most effort to try and ensure they have had a reasonable and constant credit basis (read that as one which is technically defendable under the BOINC definition of credit) even as the project has evolved over time.

The main reason for issues like Cross Project Parity problems lies in some really poor design decisions in the BOINC framework itself, a general lack of guidance from Berkeley on a proper protocol for the projects to use to determine a uniform basis to set their credit rate, and a really vocal minority which takes the position the more credit paid for work done the better (even if that position has no technical merit at all).

Alinator

Jord
Joined: 26 Jan 05
Posts: 2,952
Credit: 5,723,975
RAC: 308

RE: .. a general lack of

Message 97416 in response to message 97415

Quote:
.. a general lack of guidance from Berkeley on a proper protocol for the projects to use to determine a uniform basis to set their credit rate


That's coming. See A new system for runtime estimation and credit, although I bet that those who argue that credits are too high/credits are too low/you should get a life/no, you get a life, will not want this new system to be added.

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9,352,143
RAC: 0

RE: RE: .. a general lack

Message 97417 in response to message 97416

Quote:
Quote:
.. a general lack of guidance from Berkeley on a proper protocol for the projects to use to determine a uniform basis to set their credit rate

That's coming. See A new system for runtime estimation and credit, although I bet that those who argue that credits are too high/credits are too low/you should get a life/no, you get a life, will not want this new system to be added.

LOL...

I'll believe that when I see it! ;-)

In any event, I don't really care if the credit system is high, low, or indifferent. I just want it to be consistent, technically defendable, and enforceable. It would be nice if a system could be developed which would make it possible for the individual projects to not have to worry about setting their basis (IOW, automatic), since most science projects really don't have the time or inclination to want to deal with this issue in the first place.

You're right about one thing though, it will in all likelyhood bring about a massive period of pompous windbagging from both the hard core Kredit Kops and Credit Anarchists.

That ought to at least be good for several months of comic relief on the project forums. :-D

Alinator

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,149
Credit: 129,254,071
RAC: 13,336

RE: That's coming. See A

Message 97418 in response to message 97416

Quote:
That's coming. See A new system for runtime estimation and credit, although I bet that those who argue that credits are too high/credits are too low/you should get a life/no, you get a life, will not want this new system to be added.


Yes, LOL! :-)

Heaven forbid that we should either miss out on weighing bucketfuls of steam, or watching those that do ..... :-)

I think you're right Alinator, since Whiskymania has pretty well defined E@H to be not suitable, the solution is obvious on his/her grounds.

But the credit topic has always traditionally uncovered feelings ( occasionally of surprising vigor ) related to personal locus of control, and reciprocation/reward in a casual market. Gosh how I wish someone really well-versed in the social sciences would have a crack at explaining the BOINC milieu.

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

tullio
tullio
Joined: 22 Jan 05
Posts: 2,049
Credit: 41,465,911
RAC: 17,283

Probably the sociologists

Probably the sociologists ignore the existence of BOINC. Studies have been made on the gravitational wave community and are now being made on the LHC community. But we bear the stigma of people looking for aliens, so serious researchers ignore us.
Tullio

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1,334
Credit: 1,857,929,391
RAC: 1,505,417

RE: The reason is both ABP

Message 97420 in response to message 97415

Quote:

The reason is both ABP and GCE pay somewhat less generally speaking than GW does.

The credit ratios between projects are cpu and architecture dependent. On all 5 of my computers (4 windows, 1 linux) S5R6 is the lowest paying project, but the amount varies considerably as does if is ABP2 or GCE is the highest paying one. Whily my sample isn't big enough to draw any statistically meaningful conclusions is appears that the ratios have been adjusted to be closest to 1:1 for AMD systems, with newer intel machines scoring significantly better with the new projects than the older ones. Since all three (2 if the oldest has been retired) clusters the ligo team has running are AMD based using them as the baseline isn't surprising.

The other observation I'd make is that the first WUs for each new app are almost always granting too little credit and the number is then adjusted upward. While they always say it's because their estimates were off, the fact that they're consistantly off in the same direction probably is by design. Getting them exactly right isn't possible, and fixing them by increasing the granted credit because it's too low is less likely to make people mad than starting high and lowering it.

[pre]
credit ratios
cpu S5R6:ABP2:GCE
i7-920 1.00:1.71:1.41
s939x2 1.00:1.15:1.20
s939x2 1.00:1.20:1.20
core1d 1.00:2.17:1.83
atom 1.00:2.00:2.50
[/pre]

data for table above:
i72-920 @3.85ghz Win7-64

GCE 44 credit/hour
http://einsteinathome.org/workunit/71302782

ABP 54 credit/hour
http://einsteinathome.org/workunit/71345163

S5R6 30-32 credit/hour
http://einsteinathome.org/workunit/70315302
http://einsteinathome.org/workunit/69334315

AMD s939 dualcore @ 2ghz linux

GCE 24 credit/hour
http://einsteinathome.org/workunit/71387884

ABP2 23 credits/hour
http://einsteinathome.org/workunit/71228808

S5R6 18-22 credits/hour
http://einsteinathome.org/workunit/70255521
http://einsteinathome.org/workunit/69423990
http://einsteinathome.org/workunit/70263352

Intel Core1 Duo at 1.73ghz Win Vista

GCE 22 credit/hour
http://einsteinathome.org/workunit/71387957

ABP2 26 credit/hour
http://einsteinathome.org/workunit/70406439

S5R6 12 credit/hour
http://einsteinathome.org/workunit/71071965

Intel Atom @ 1.6 Win7

GCE 10 credits/hour (after adjustment - this machine has only got an old 160 credit WU, not any new 250 credit WUs)
http://einsteinathome.org/workunit/71211420

ABP2 6-8 credits/hour
http://einsteinathome.org/workunit/70226813
http://einsteinathome.org/workunit/70871766

S5R6 3.5-4 credits/hour
http://einsteinathome.org/workunit/70733209
http://einsteinathome.org/workunit/70786208
http://einsteinathome.org/workunit/70845140

AMD s939 dual core @ 2.4ghz WinXP

GCE 24 credits/hour
http://einsteinathome.org/workunit/71392342

ABP 24 credits/hour
http://einsteinathome.org/workunit/71228808

S5R6 19-21 credits/hour
http://einsteinathome.org/workunit/70255521
http://einsteinathome.org/workunit/71030903
http://einsteinathome.org/workunit/71005416

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.