BOINC, Burp, and Benchmarks......Oh My!

[CarcinogenX]

So, I’ve read up on BOINC and optimizing a little but it appears that I am still doing something wrong.
First, I understand that this is an alpha project, but others are having a much better go at it than me. Looking at my results, it’s clear that I am getting shafted most of the time. I spend hours on a project, claim say, 200 credits, but am granted far, far less. While the other guy and his 300mhz Celeron fart out 20 minutes of cpu time, claims 5 credits and get 100. It’s a bit of an exaggeration but what gives??
I noticed that Loki has a somewhat similar CPU to mine and looking at his results, he generally receives what he is granted and sometimes far more. I’m running just plain old BOINC 10? Should I be doing something else?
One more thing I don’t get. My system is circa 2005/6 and when I run a benchmark, I get higher numbers than most computers out there. Why is that happening?
My current results are:
2885
5223
Xpod has a C2D E6850 that gets lower benchmarks than me. I know that cannot be correct. Should I be setting my multiplier to 1x before running a benchmark on a project?

Any help would be much appreciated.

[mitchellds]

I feel your pain. I have the same problems. Here is one of my pc's showing credit claimed and granted. I could only find one result that they gave me more credit that I asked for. I'd say that 95% of the credit returns less than 30% of was was claimed. I wonder what causes this poor estimation? Is it the project applications, or the way boinc estimates wu completion/credit?

Maybe someone can post to the project admin for answers?

240.69 122.79
247.98 70.04
218.91 83.36
144.40 37.34
259.17 70.73
163.14 33.53
227.19 51.81
126.60 50.24
217.80 48.95
167.25 66.53
155.29 41.32
240.69 122.79
230.46 74.59
207.77 121.93

0.94 1.01

[NeoGen]

Burp grants the middle claim out of the first three that arrive for each workunit.
Say for instance three people return a result for a workunit, and the first claims 5, the second claims 10, and the third claims 15. Actually, it's independent from the order by which those three arrive, everyone is given 10 as soon as the third result comes in.
This is for protection against people with optimized boincs that make huge claims, but since that sometimes new benchmark values are achieved with different (stock) boinc versions on the same machine, it is indeed annoying to not get what we claim.

Guess it's time for them to revise their credits strategy...


P.S. - Also... Welcome to the forums CarcinogenX

[Strongbow]

In addition and one that I really have a gripe about is regarding the average work done rating! ...I can be crunching their day+ WUs and my average drops off pretty much immediately, hence the reason why it is hard to actually get a realistic figure to the amount of daily points you gain!

[Bender10]

This is an often debated problem in BOINC projects. I may not be able to nail this down for you, as my information comes from the "www.apsathome.org" site (which seems to be down). If the site comes back up, check there. Bear with me during my short blurb.

Also, Alpha stage projects (and the credit for WU's) tend to be a little quirky for the testers... The designer for BURP is hopefully working on what they call 'Equalizing' the amount of credit awarded, or how is is awarded.

Remember, you are in the mystery world of 'Opimized Clients', which through smoke and majik, provide higher point per WU's than normal clients (most times). When 'Optimized Clients' don't werk, I like to think the project designer did a good job designing his points algorithm.

And BURP uses a 'Qorum of 3' to award points. For that, They send the WU to 3 computers and wait for the results.

EDIT: Ooops. They try to get 3 cpu's to agree on a score. If that does not happen, the WU keeps getting sent out. Then they filter for errors/totals/successful results. Kinda complicated. see below.

application Blender
created 23 May 2007 16:33:33 UTC
name 450in0.zip__ses0000000450_frm0000000209_prt00025.w u
minimum quorum 3
initial replication 4
max # of error/total/success results 25, 35, 20

Click on a "Workunit ID" number, and you will see the above information. This will tell you the critiera being used to find 'successful' matching results.

Then they normally throw out the high and the low and award all computers the middle value (in a nutshell).... So yes, a slow machine can get more points than they have claimed for a specific WU. Kinda squirrely...

(most crunchers don't like Qorums anyway...).

Any comments? or corrections?

Back to werk I go....

[Lagu]

CarinogenX!

Welcome to our forum! I hope you can got an answer on your dilemma. I´m nor the right person for your problem bat I´m glad se a new "face" into our forum. Are you crunching for AMDusers?

Lagu

[CarcinogenX]

CarinogenX!

Welcome to our forum! I hope you can got an answer on your dilemma. I´m nor the right person for your problem bat I´m glad se a new "face" into our forum. Are you crunching for AMDusers?

Lagu

Thanks Lagu,

I'm providing whatever help I can with the Burp "race" for the AMDusers team.
I'm still need a little clarification though.
I get an 8hour WU that, through a nasty vortex of space/time takes maybe 12+ hours to complete.
The "other" guy and his 300mhz Celeron must be getting a considerably smaller WU than me, because it is completed in the same or less time than me.
So I don't get the connection that I and the Celeron guy have that should grant us the same credit.
And I'm not here to bitch or complain, I am curious and just want to understand the process better.
Also, someone else mentioned it too that the long WU's kill your RAC. It seems like it's based on a very short time frame, shorter than other projects.
Nobody mentioned anything about benchmarks either. I don't know how much of a role they play and why mine are so high. I read that some optimizers will beef up your benchmarks, but I am only using the latest version of BOINC.

[CarcinogenX]

I think the real answer is for all of us to switch to really old Sempron's

[Lagu]

I too run BURP on my Intel Core 2 Duo 6600, 2,4 GHz and I earn relatively good points. I began crunch July 7 and have now 4,000 points.

Here is my Boinc Benchmark:
2007-07-25 01:59:20|| Number of CPUs: 2
2007-07-25 01:59:20|| 2240 floating point MIPS (Whetstone) per CPU
2007-07-25 01:59:20|| 4691 integer MIPS (Dhrystone) per CPU

I have noticed BURP has his own Benchmark as chows here:

Measured floating point speed: 2,244,75 million ops/sec
Measured Integer speed: 2,992,23 million ops/sec

That is strange. What is valid? BURP or Boinc or both? On other project Boinc benchmark is the same on the projects page but not here?

Example of CPU-sec, Claimed and Granted credit:
34792, 139,89, 117,19
27718, 84,0, 84,0
27800, 84,25, 113,85
27784, 91,61, 126,04

I received mostly WU´s as takes between 4,5 to 6 hours. I also only receive 2 WU´s per time, crunch them and get 2 new the whole time.

On my 3 other computers I run MoneyBee. Have you an older Intel PII you can run MoneyBee on this computer.

I haven’t owe clocked any of my computers because we have summer in Sweden and the computer room gets too hot.

[AMDave]

Excellent post there Nflight!

I don't see much BURP dissent. There are some firm opinions from those who prefer more stable projects which is understandable as BURP is an Alpha project, but allowing for the rough edges of an Alpha project, BURP does alright in the grande scheme of things.

As far as AMD advantage goes, I still cannot go past Riesel Sieve Classic (Unless you have a Mac-G5 using JoeO's custom compiled client)

As far as BOINC goes I have found the current BURP credit awards to be comparable with some other projects for my middle-of-the-market CPUs [EDIT] on closer inspection some CPUs do seem to be disadvantaged on BURP currently [/EDIT]. Otherwise RieselBOINC is giving some decent credit and uFluids is on the kinder side as well.

It simply depends on your CPU type and available RAM. If you have and AMD64 based CPU (either single core or SMP) then you want to head for the BOINC projects that have specially compiled clients for AMD64 rather than those projects that simply allow the server to send a 32-bit client to an AMD64 CPU.

Augustine has been following (and championing) this 32-bit to 64-bit adoption and you can follow his updates on the BOINCstats forum here
http://www.boincstats.com/forum/foru...ad.php?id=1774

I will not run a system under 2GB RAM anymore so I don't tend to hit any RAM issues very often although I run a lot of services on my box that I run linux on so every once in a while it rejects a WU that needs more RAM than what is available, but the other boxen just plod along now.