Advanced search

Forums : General Topics : Cosmology@Home credit discussion
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 13 · Next

AuthorMessage
The Gas Giant

Send message
Joined: 12 Dec 07
Posts: 5
Credit: 15,100
RAC: 0
Message 4413 - Posted: 7 Jan 2008, 20:21:23 UTC

Some you win on, some you loose on. All up it\'s pretty close to the optimised SETI app for overall credit. I get 100cs for 11,000 seconds and I get 100cs for 23,000 to 30,000 seconds of cpu time, to give a daily RAC of somewhere between 550 and 1000 for my old 3.0GHz P4 Prescott. With the optimised SETI app this machine has a RAC near 800 to 1000 and on Malaria Control which doesn\'t have an \"optimised\" app, this machine gets a RAC of between 350 to 400.

I must say that I\'ve not let this machine run Cosmo at 100% resource share to get a true daily figure, the numbers above are based around the wu\'s it\'s completed in the last 3 weeks.

Glad I\'m not a credit whore otherwise a very worthwhile project wouldn\'t get any of my CPU time......

Live long and BOINC!

Paul.
ID: 4413 · Report as offensive
transient

Send message
Joined: 7 Oct 07
Posts: 7
Credit: 109,450
RAC: 0
Message 4418 - Posted: 8 Jan 2008, 14:33:24 UTC

I\'m running Seti(optimized app) and Cosmology on the same box with the same resource shares. The last time I checked, a couple of weeks ago, they had a similar (BOINCStats)-RAC.

It is not interesting enough for me to keep a close watch. ;)
ID: 4418 · Report as offensive
Profile Angus

Send message
Joined: 28 Jul 07
Posts: 71
Credit: 31,250
RAC: 0
Message 4419 - Posted: 9 Jan 2008, 4:31:35 UTC
Last modified: 9 Jan 2008, 4:32:14 UTC

Let the credit wars begin!

Posted today on boinc_projects mailing list:
There is significant variation in average credit
per CPU second between projects: see
http://boinc.netsoft-online.com/e107_plugins/boinc/get_cpcs.php

The following projects are giving more than 1.5X S at h\'s credit/sec:
QMC (1.59)
RieselSieve (4.3)
Cosmology at home (1.68)

... and a number of projects are giving somewhat less (in the .7 range).

It is critical to the success of volunteer computing that participation
be based on factors like scientific merit - not points per CPU sec.
So can all projects please take steps to make their ratio
as close to 1 as possible (and no more than 1.2).
If not, I will eventually ask the statistics sites to scale down
the credit of projects with high ratios.

To adjust your ratio:
1) if you\'re giving fixed credit per WU,
or using the FLOPs-reporting API, change your numbers.

2) if you\'re giving credit based on benchmarks*CPU time (the default)
set the <fp_benchmark_weight> configuration parameter
to a value that\'s appropriate for your application: see
http://boinc.berkeley.edu/trac/wiki/ProjectOptions
If you\'re not sure how to do this, or it doesn\'t fix things,
please reply to this group.

-- David
ID: 4419 · Report as offensive
Profile kevint

Send message
Joined: 30 Aug 07
Posts: 46
Credit: 6,502,980
RAC: 0
Message 4420 - Posted: 9 Jan 2008, 4:45:18 UTC - in response to Message 4419.  


Always got to be someone to piss in the pool so to speak.

Is SETI feeling that it may be loosing contributors because of it\'s tightened down credit ?

And Who is this \"David\" Is it David Anderson?

If that graph is based upon plain optimized SETI it might make more sense.

As a general rule, those who care about credits are going to optimize, those that don\'t, won\'t.

So this chart
http://boinc.netsoft-online.com/e107_plugins/boinc/get_cpcs.php
should be corrected to baseline with optimized SETI IMO.

ID: 4420 · Report as offensive
Profile Angus

Send message
Joined: 28 Jul 07
Posts: 71
Credit: 31,250
RAC: 0
Message 4421 - Posted: 9 Jan 2008, 5:35:21 UTC


And Who is this \"David\" Is it David Anderson?


Yes, David Anderson, Dr. BOINC.
ID: 4421 · Report as offensive
transient

Send message
Joined: 7 Oct 07
Posts: 7
Credit: 109,450
RAC: 0
Message 4422 - Posted: 9 Jan 2008, 8:55:26 UTC - in response to Message 4420.  


If that graph is based upon plain optimized SETI it might make more sense.

As a general rule, those who care about credits are going to optimize, those that don\'t, won\'t.

So this chart
http://boinc.netsoft-online.com/e107_plugins/boinc/get_cpcs.php
should be corrected to baseline with optimized SETI IMO.



I think the graph does not make a distiction between optimized and not optimized applications. Either way, the number of optimized apps versus the number of not optimized apps is bound to be low.
ID: 4422 · Report as offensive
Profile Jayargh
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 25 Jun 07
Posts: 508
Credit: 2,282,158
RAC: 0
Message 4423 - Posted: 9 Jan 2008, 15:43:24 UTC - in response to Message 4422.  
Last modified: 9 Jan 2008, 17:13:51 UTC


If that graph is based upon plain optimized SETI it might make more sense.

As a general rule, those who care about credits are going to optimize, those that don\'t, won\'t.

So this chart
http://boinc.netsoft-online.com/e107_plugins/boinc/get_cpcs.php
should be corrected to baseline with optimized SETI IMO.



I think the graph does not make a distiction between optimized and not optimized applications. Either way, the number of optimized apps versus the number of not optimized apps is bound to be low.


A few points here.....

Though the # of optimized apps used at Seti may be lower than the non,the # of results crunched is orders of magnitude higher than the non....in other words active crunchers optimize.

The referenced chart is just that a reference...I don\'t understand why scientists at Boinc are using this as a bible? That list is missing multiple Boinc projects such as LHC and RCN as quick examples....why are they only targeting the 3 projects QMC,Riesel Sieve,and Cosmology when that list is not comprehensive? When was the last update of that list? I don\'t see that either.I am getting more credits per hr in Milkyway than in Cosmo but the chart says I should be getting more than 50% greater in Cosmo so how accurate is using the chart for comparasins?

How do they propose to deal with PS3\'s? Obviously they produce orders of magnitude greater credit than pc\'s and they are not targeted or even mentioned by the Boinc staff.....parity is a dream because of all these factors and they should at least include all Boinc projects not just the ones on an outdated list.

I see team members crunch just Seti optimized just to keep up in Boinc totals to those who crunch Seti-only...caused by the very project they are asking to have parity to taking away from other projects.....perhaps they should eliminate optimized apps at Seti 1st before they make the call for other projects to change their credit granting.,if they really want to solve this.

The final paragraph says they will ask the stats sites to change the credit if the projects won\'t.....coming from someone with a degree and experience dealing with people this is a very odd soloution because some stats sites may not agree to do this and in so doing credit will be different everywhere making credit more meaningless.

You would have thought Dr.Anderson would have thought this out a little better than what I have read here.
ID: 4423 · Report as offensive
Soriak

Send message
Joined: 24 Aug 07
Posts: 17
Credit: 177,330
RAC: 0
Message 4424 - Posted: 9 Jan 2008, 16:20:40 UTC
Last modified: 9 Jan 2008, 16:22:27 UTC

The numbers seem to come from boincstats\' chart comparing the credits of hosts that work on both projects.

This only seems like a rudimentary guideline though, it does not account for projects performing differently depending on the processor architecture. Certain optimization may give a much bigger boost to SSE3 processors and if the credits are benchmarked on one without SSE3, the hosts that have it will receive above average credits. On the other hand if it\'s benchmarked on one with SSE3, everyone else will receive much less credit. Some projects also run better on AMDs than Intels or the other way around.

Also fails to consider alpha/beta projects that grant more points because of complications that come up. The WUs that don\'t give any credit don\'t show up in such a comparison chart but do affect the long-term credit one receives.


The worry that people will gravitate towards projects that give a lot of credits is - at least so far - not supported by evidence. We\'d see a vast majority of people crunching RieselSieve and some others crunching Cosmology, if that were true. Even if we assume there are different groups of interest - ie \"math\", \"space\", \"biology\" and so on - we\'d see a shift towards the projects with a lot of points within that group.

People, however, aren\'t doing that: Cosmology gives 1.9x the credits that Einstein does, but Einstein has 18x the number of users and 48x the number of active hosts.

RieselSieve, by that measure the far most generous project, has 2,300 active users (same as cosmology) - compared to the 200,000 active users in SETI and SETI beta. The number isn\'t a whole lot different for hosts eithers - Cosmology has 9,000 (Riesel about 21,000), SETI has over 1,700,000.


So the data really doesn\'t support a shift towards project that give comparatively a lot of credit, hence questioning how much people care about the BOINC combined statistics. I think this really is a non-issue except to a few vocal people.
ID: 4424 · Report as offensive
Profile Jayargh
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 25 Jun 07
Posts: 508
Credit: 2,282,158
RAC: 0
Message 4426 - Posted: 9 Jan 2008, 18:20:07 UTC
Last modified: 9 Jan 2008, 23:25:22 UTC

I would also agree with which some have posted before me saying the Seti-standard ought to be the optimized.

If Boinc devs are truly interested in parity the will realize that they are open source which most projects are not for many various reasons ...the most common being the licensing of the app from a 3rd party which will never be open source and hence will probably never be able to be optimized.

This recognition by the Boinc deveopers would help level the playing field as new projects feel the pressure to keep up with Seti optimized.

In my opinion there should be no 3rd party optimized apps....only internal...either everyone of that project participates or no-one. This type situation almost jeopardized Einsteins validaty in their peer review until they took them internal.Have 3rd party apps be submitted to the project for internal optimizations.Seti has caused the problem and needs to fix the problem before witch hunts are started.

They opened the Pandora\'s box of credit and now wish to force a truly bad fix using questionable data.
ID: 4426 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4429 - Posted: 9 Jan 2008, 18:47:42 UTC - in response to Message 4426.  
Last modified: 9 Jan 2008, 18:54:54 UTC

I would also agree with which some have posted before me saying the Seti-standard ought to be the optimized.

If Boinc devs are truly interested in parity the will realize that they are open source which most projects are not for many various reasons ...the most common being the licensing of the app from a 3rd party which will never be open source and hence will probably never be able to be optimized.

This recognition by the Boinc deveopers would help level the playing field as new projects feel the pressure to keep up with Seti optimized.

They opened the Pandora\'s box of credit and now wish to force a truly bad fix using questionable data.


SETI reduced their credits in August, when the Multibeam application came out. The credit multiplier was reduced from 3.35 to 2.85. From what I can tell with my hosts, since I was using the optimized SETI app, my overall credit ratio is about the same here as it was/is there. Sure, I may get some that are higher here, but I also get some that are higher there. All in all, maybe Cosmology is offering 1.1-1.2x...when compared against the optimized application.

As for SETI specifically, they already cannot adequately support the current number of participants they have. They already have serious server problems every week, if not every day. They have a large amount of data that has to be gone through, so they need more participants, but in order to keep those participants happy, they need to improve their server infrastructure. They would say \"donate money or donate hardware to help\", but some of us simply can\'t. In my specific case, I\'m unemployed (layoff...and a very \"dead-end\" IT town, but I don\'t have the money to move)...

I don\'t know where all of this is going to lead, but it could be signaling an end to the open source portion of SETI\'s science application.

Oh, and as to \"scientific merit\", more and more I am seeing no point to doing SETI. They are not doing post-processing of the results that have been processed. They want to do post processing, with what\'s called Near-Time Persistency Checker (NTPCKR, aka \"nitpicker\"), but they currently have no hardware dedicated to it and are trying to cobble together a proof-of-concept system that will be significantly underpowered for the task. The person that heads up the hardware donations over there also is looking for something that could indeed be underpowered, but that is \"affordable\", although I question the wisdom of doing so. They seemed to be scraping for the bottom-end solution, when you should at least target the bottom of the midrange.

ID: 4429 · Report as offensive
Profile Jord
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 15 Jun 07
Posts: 345
Credit: 50,500
RAC: 0
Message 4433 - Posted: 9 Jan 2008, 22:56:34 UTC

There\'s a new [trac]wiki:CreditProposal[/trac] that everyone can read. Needs some understanding of higher mathematics, though.
ID: 4433 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4434 - Posted: 9 Jan 2008, 23:19:11 UTC - in response to Message 4433.  

There\'s a new [trac]wiki:CreditProposal[/trac] that everyone can read. Needs some understanding of higher mathematics, though.



Using the published credit functions, it would be possible to develop a \"credit maximizer\" web site, where users can enter the parameters of their computer, and it tells them how much credit/day each project would give them.


I thought the point of all this was to flat out normalize credits across projects? If so, then there\'d be no point to a \"credit maximizer\" type thing like what\'s being proposed.

Personally, I think David Anderson should quit worrying about this. I haven\'t cut back on my SETI participation because Cosmology is offering \"more\" credits, because they aren\'t offering that much more, if any, because I am using the optimized SETI application and not the stock application. I have cut back my participation because of his tinkering with the forum/website code, the fact that there is no post-processing, the continual server problems that show that they can\'t support all the users that they have, and the \"oh well, it will pass\" attitude of Matt, who also seems to not understand that he SHOULD care about the deadlines because they are a contributing factor to his ever-increasing need for uploaded result storage space. However, he\'d rather dismiss it as \"not my area\", and while I understand that point of view, it shows that either he doesn\'t understand all the dynamics, despite claiming to be one of the \"players\" on the \"basketball team\", or he just views it as noise by complainers of having excessive pending credit, which means he doesn\'t take the time to fully read the content of the post before dismissing it...
ID: 4434 · Report as offensive
Profile Beezlebub

Send message
Joined: 11 Aug 07
Posts: 63
Credit: 1,843,380
RAC: 0
Message 4441 - Posted: 10 Jan 2008, 17:25:12 UTC

I quit running Seti sept, 2007 again because of the server problems and downtime they were having. I had started with Seti beta in 1999 got tired of all the problems after a couple of years. As Brian says, David etc. need to fix Seti FIRST then worry about what other projects are doing. (with MUCH less trouble)
ID: 4441 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4443 - Posted: 10 Jan 2008, 20:31:22 UTC - in response to Message 4441.  

I quit running Seti sept, 2007 again because of the server problems and downtime they were having. I had started with Seti beta in 1999 got tired of all the problems after a couple of years. As Brian says, David etc. need to fix Seti FIRST then worry about what other projects are doing. (with MUCH less trouble)


In fairness, as you move up in the size of the project, the task of managing it becomes more difficult. The database I/O becomes much heavier, thus there is more of a need for DBA activities, such as index rebuilds, query tweaks, etc... Smaller projects have an easier task from a database management standpoint...

ID: 4443 · Report as offensive
Profile ohiomike
Avatar

Send message
Joined: 17 Jul 07
Posts: 302
Credit: 5,006,319
RAC: 0
Message 4444 - Posted: 11 Jan 2008, 1:59:14 UTC
Last modified: 11 Jan 2008, 2:00:31 UTC

It is curious that this shows up just as SETI also announces that they need more volunteers (too many going to other projects?).
PS- My numbers here match up about 1 to 1 with what I get running SETI optimized apps.


Boinc Button Abuser In Training >My Shrubbers<
ID: 4444 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4446 - Posted: 11 Jan 2008, 5:56:23 UTC - in response to Message 4444.  
Last modified: 11 Jan 2008, 6:10:27 UTC

It is curious that this shows up just as SETI also announces that they need more volunteers (too many going to other projects?).
PS- My numbers here match up about 1 to 1 with what I get running SETI optimized apps.


I\'m thinking it depends on the architecture as to whether it\'s 1:1 or not... I think my Northwood Pentium 4 2.4B (533FSB) system is doing better here than it does with SETI, but my San Diego core AMD Athlon 64 3700+ (overclocked to 2750MHz, which also gives DDR-500, so roughly FX-57 performance), seems to be doing about the same... I\'ll try to see if I can get some comparison data on it... BOINCstats cr/sec is probably a fair thing to use for my Intel system, as it has almost 100% use of 2.2/2.2B/2.4, where my AMD system has been attached for much longer and has gone through overclocking \"upgrades\" (I used to have a 3200+ overclocked to 2500MHz @ 250x10)

Edit:
These are the figures for my Intel system:
Average credit per CPU second 0.008551 (Cosmology)
Average credit per CPU second 0.003710 (SETI)

So, yes, there is a large difference there...

My AMD shows this:
Average credit per CPU second 0.008813 (Cosmology)
Average credit per CPU second 0.006535 (SETI)

Much closer there, however the SETI figure has years of historical data behind it where I didn\'t use an optimized application, so I\'d guess it is actually closer than what those figures would indicate...

Also, there are two ways to spin that information on my Pentium 4. First, one could claim that Cosmology was granting far too many credits. The other way to look at it though is that you could say that SETI is inefficient with my specific Pentium 4 system...

:shrug:

Brian
ID: 4446 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4447 - Posted: 11 Jan 2008, 7:04:49 UTC - in response to Message 4446.  
Last modified: 11 Jan 2008, 7:13:45 UTC


These are the figures for my Intel system:
Average credit per CPU second 0.008551 (Cosmology)
Average credit per CPU second 0.003710 (SETI)

So, yes, there is a large difference there...

My AMD shows this:
Average credit per CPU second 0.008813 (Cosmology)
Average credit per CPU second 0.006535 (SETI)



I\'ve been scratching my head on this and I have no idea why my Intel system shows that close to my AMD system based on that stat from BOINCstats... Simply glancing through task lists on both systems seems to invalidate the idea that they are that close. I want to say that it is likely that the Intel stat is significantly overstated on BOINCstats and/or I don\'t fully understand what that stat is trying to say. My AMD system can approach double the RAC over at SETI, which is supported by the stat figures. There is no way that the Pentium 4 could be that close. I should see the ~2X difference here too...

:scratches head:

Edit: A-ha! I think I know... I think it factors in the amount of credit somehow. Since there are fixed credits here, I\'d need to do more work on my AMD, I think... I use my AMD to do SETI, Cosmology, LHC, and Einstein, including some dabbling with some Linux virtual machines over at Einstein, where my Intel system is split 80/20 (now 85/15) between Cosmology and SETI.
Sooo, let\'s test the theory... I\'ll dedicate the AMD to Cosmology for a couple of days and see if the stat goes up...
ID: 4447 · Report as offensive
Dagorath

Send message
Joined: 24 Aug 07
Posts: 9
Credit: 35,300
RAC: 0
Message 4449 - Posted: 11 Jan 2008, 17:36:11 UTC - in response to Message 4433.  

There\'s a new [trac]wiki:CreditProposal[/trac] that everyone can read. Needs some understanding of higher mathematics, though.


Higher maths or maybe just half-baked maths? I mean the maths seemed a little difficult at first but then I realised it\'s just because there are gaps that need to be filled in before it can be turned into code. The gaps are of course necessary because each project will need to fill in the gaps in a way that suits their project. I doubt there is a general solution that fits all projects.

I am not sure who wrote the proposal but if Berkeley ever implements the proposal server side then I imagine many projects will issue the standard excuses for not adopting the new server code (as they should)...

1. when we get time we\'ll examine the proposal
2. (6 months later) we are currently reviewing the proposal
3. (6 months later) we are currently looking at the new server code but it isn\'t easy, we hope Berkeley will provide a credit specific API or something to make it easier to implement, we encourage anyone who can code and has spare time to help write the API
4. (6 months later) we have a few scripts and a new science app that might implement the new credit system and we might be ready to test them next year
5. (next year) we lost the work in a recent disk crash so we\'ll have to start over, sorry there was a fault in our backup script but it\'s fixed now ;)
6. we are happy to announce that we have concluded our investigation, this project will now shut down, thank you all for your devotion and hard work, results will be published soon, btw, we are deleting all our work on the new credit proposal because the problem the proposal seeks to rectify (crunchers are gravitating to projects that pay high credits) simply doesn\'t exist

ID: 4449 · Report as offensive
XSDiabloLegion

Send message
Joined: 22 Dec 07
Posts: 3
Credit: 13,900
RAC: 0
Message 4450 - Posted: 12 Jan 2008, 0:11:45 UTC

Honstly: The whole issue of cross comaprig credits is the pits and is designed to create havock in the sleeping patterns of other project managers. Ignore
ID: 4450 · Report as offensive
Brian Silvers

Send message
Joined: 11 Dec 07
Posts: 420
Credit: 270,580
RAC: 0
Message 4454 - Posted: 13 Jan 2008, 19:22:56 UTC - in response to Message 4447.  
Last modified: 13 Jan 2008, 19:27:06 UTC


These are the figures for my Intel system:
Average credit per CPU second 0.008551 (Cosmology)
Average credit per CPU second 0.003710 (SETI)

So, yes, there is a large difference there...

My AMD shows this:
Average credit per CPU second 0.008813 (Cosmology)
Average credit per CPU second 0.006535 (SETI)



I\'ve been scratching my head on this and I have no idea why my Intel system shows that close to my AMD system based on that stat from BOINCstats... Simply glancing through task lists on both systems seems to invalidate the idea that they are that close. I want to say that it is likely that the Intel stat is significantly overstated on BOINCstats and/or I don\'t fully understand what that stat is trying to say. My AMD system can approach double the RAC over at SETI, which is supported by the stat figures. There is no way that the Pentium 4 could be that close. I should see the ~2X difference here too...

:scratches head:

Edit: A-ha! I think I know... I think it factors in the amount of credit somehow. Since there are fixed credits here, I\'d need to do more work on my AMD, I think...
<snip>
I\'ll dedicate the AMD to Cosmology for a couple of days and see if the stat goes up...


My theory, or at least some aspect of it, has indeed proved correct. My credit per cpu second for my AMD system here at Cosmology has increased, while the same stat for my Intel has remained roughly the same...

Average credit per CPU second 0.011226 (AMD)
Average credit per CPU second 0.008523 (Intel)

Now, as to the wide spread on SETI vs. Cosmology on my AMD... Several factors at play there... First, the system was originally a 3200+ with 1GB of memory instead of a 3700+ with 2GB memory. The 3200+ processor only had 512k of cache, while the 3700+ has 1MB of cache. SETI has shown to be responsive to a cache increase, but with the biggest \"bang for the buck\" going from 512k to 1MB... Additionally, my overclock on the 3200+ was only up to 2500MHz, where as now I have the 3700+ at 2750MHz... I\'m pretty sure that I also ran an unoptimized application for a while over there, so the data has in it some results that took longer... Also, as noted here, running results seems to influence that particular statistic, and since I\'ve run a few SETI results since posting the earlier message, here\'s the current stat for the AMD at SETI:

Average credit per CPU second 0.006535 (original post)
Average credit per CPU second 0.007342 (now)

Thinking about this, I didn\'t run all that many results over at SETI over the past few days. I think I should probably get in touch with Willy (the guy that runs BOINCstats) and ask exactly what the calculation is for that stat and how much history is involved in the calculation...
ID: 4454 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 13 · Next

Forums : General Topics : Cosmology@Home credit discussion