I know that the know it alls will try to explain this away as some problem other that the new and improved credit system. But the Facts Remain Facts, and numbers don't lie. This has to be the lowest PPD project in Boinc right now.

My RAC has been steadily dropping since 12/06/06.


Img tags don't seem to be working so here is a direct link.
http://img155.imageshack.us/my.php?image=racdroppingqx1.jpg

I was averaging 1320-1340 PPD, now as you can see it's down to 1237 and still dropping, with no end in sight.

I an not effected by the recent errors that many have encountered. I have had no more or less than before.

So the question is How much has your RAC dropped?

I won't try to explain it away, there's not much point of trying to do so with your computers hidden. Mine has dropped in that timeframe as well, and I can explain essentially all of it on the BOINC manager losing contact with the client, the screensaver lockups, Ralph sending tasks out in that timeframe (so time crunching away from R@H), and a squirrel that found a creative way to combine a power transformer with his incessant desire to explore the afterlife.

The graph of credit issued by the entire project each day would seem to show a measurable decline in about that same timeframe.

If you run down your list of tasks, how does your claimed compare to granted? I mean really the claimed credit is the old credit system, and the granted is the new, so that's what you want to be studying. If you see some granted = 20, you may find that the watchdog stepped in and successfully ended tasks "normally", but it really ended because the watch dog felt it wasn't proceeding properly and these only seem to get 20 credits, regardless of how much time was in to it.

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

I was right, tying to explain it away. Don't blame it on the new credit system, it has to be something on your end.

Let me try this one more time.

I have 4 machines, they only run Rosetta 24/7. My 28 day average was 1333 on the 6th. Right now its 1272, and my 7 day average is 1198. I have had 3 errors 1 each on 3 different machines over this time frame. 1 was a wu that had already been ran, and came back with an error that stated too many results =20 pts. 2 errors were video errors =20 pts.

Like I said not an over abundant amount of errors. I don't use the screen saver, or look at the graphics while its running. I have not had any communication problems. Just the granted credit has been steadily declining, claimed credit has remained the same. Each machine has been dropping the same amount, over 25 pts each per day since the 6th.

So you see it's not on my end. It's the credit system, and something is wrong. I think the graph that you pointed to backs up my facts.

Why can't you just admit that there is a problem with the credit system. I would like to see that charted cross platform parity spreadsheet now to see just where this project is in terms of ppd compared to the other projects now. I'd bet it's at the bottom, or lower.

Nice way to attract new members. Come on we will give you less credit than anyone else.

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

-_-

---

-_-

Nice, always one Axxxxxx.

To try start Sxxx.

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

Stevea, problem is with you computers hidden it is hard to see what you got.

Feet1st asked a good question there though.

Add up all the claimed credits and compare it to the granted credit.
You will then get an which computers maybe dropping the most. (i.e. is this compile of rosetta@home slightly slower or faster on a processor, since yours are hidden we cannot do it for you)

As far as we know, the failed tasks will bring down the rolling average of credit and hence bring down everyones credit, that is what is probably wrong. Nothing we can do about it and certainly not till the new year when the project staff are back from the holidays.


How much have mine dropped, no idea.
The Pentium-M is still granted more than claimed. (same as always)
The Pentium4-m sometimes claims more, but more often than not is is less than claimed. (same as before)
The Pentium3-m is about where it was before.
The A64, I don't know it's not running very often at the moment.
The AXP I also don't know, I think it's about the same as normal.
I will check (or you can look at mine for yourself).


We'll need Asto for the project comparisson graphs, but it christmas and he's also busy with ABC@home.


To rule out stats site collection problems (your's looks like boks FDC)
Here's yours at boincstats.

---

Team mauisun.org

LOL, I'm still around. I spent Christmas with the folks in Michigan, just got back(South Carolina). Ran Rieselsieve, Chess960, Proteins, Spinhenge, and Ufluids while I was gone. Set it to high cache of 3days (usually 12 hours) to carry these through any outage. It'll be a few before I can do anything.

I've done some Rosetta a week or two ago, and it appears something happened to lower granted credit a snidge, but I can't swear to it. Nothing appeared "Drastic" IMO, so I was unconcerned.

Well, back to my Newegg shopping page. lol I see they're offering the Socket F FX-74 for $550 bucks,, hmmm

---

I wasn't sure if Steve was going to retain the attitude through this thread or if it was just poor word choice on his initial post.

Yes Steve! I INTENDED for the chart I sited to SUPPORT your assertion that "something" happened about a week after the date you sited. As for numbers not lying and etc. the chart I sited PROVES absolutely nothing, because there is no way to know what the other 50,000 crunchers were actually doing during that time. And the several thousand "new hosts" as well.

On MY end, when I review all available tasks for my three machines I see that my claimed credit and granted credit are basically in-line with one another. I see no decline in the ratio between the two during that timeframe (although it doesn't go back to your date). Check them your self. In fact, I was sometimes granted significantly MORE credit then I claimed.

I'd be happy to "admit" that a system is broken, if only someone would point me to something specific which demonstraits that. Third-party stats are always a little suspect. RAC is too broad a measure. Credits never have had true equivelance between projects, and that was not the objective on the new credit system.

So, if you wish to claim that the new credit system suddenly broke several months after it was rolled out, your best argument would be to point to some specific tasks that did not grant proper credit, or to a machine that is suddenly bringing down your average. Without that, you've done nothing to help resolve the problem, nor to create a better credit system, nor to improve the project. You've simply voice an opinion that the value of Rosetta credits should be inflated to some level that suits you.

I would be glad to help you support your claim in order to help investigate and potentially address the cause. Since I don't count credits for much of anything, I for one am not well versed in the third-party stats. Please outline the steps one should follow to display their nice graph of all their machines consolidated in to one number, and if possible, include how to also show Ralph credit during the same timeframe, and to find the WUs that I had in that timeframe that were only awarded 20 credits due to the watchdog (and I already agree, that is one specific area that could/should be reviewed/improved, but it doesn't seem to be your point here).

In short "how do I find the answer to the question posed in the subject of this thread?"

In case anyone is inclined to assemble such a graph for me, here are my host IDs:

Rosetta Ralph
271072 4725
172896 1640
181314 4737

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

I'm not here to stoke any fires, but more than one person has observed and reported a drop in RAC "recently". Inquiring minds want to know why. In particular we wonder if this is real or an abberation, whether temporary or permanent. If real, what is the cause?
As I reported here , even after increasing my resource share to Rosetta my RAC continued to drop in December. For me it's real and seems permanent, though I have no supporting evidence other than my own observation. Since credits don't matter in the real world I don't feel compelled to do more than say that more than one person has observed a drop in RAC.

Not taking sides, but I thought I would throw out a number here.

I have to many results to sort through for December. If anyone wants to do it, feel free. But, for the first few pages of my results, I claimed 2474.19 and was granted 2505.56.

My computers are available to look at, and since I have so many results this month, it would be a good section of results.

The only thing I am worried about is that I am getting to play with some nice blade servers right now, and people like me could be bringing down the average results since the systems I am using are higher ended Duel Proc servers.

Does anyone know if computer type is taken into account when awarding credit? If a brand new duel or quad core 4 proc server is compaired to your average home computer system, it doesn't really seem fair.

I haven't noticed much variation on my computers (mostly Athlon XP running linux).
Here's some graphs. Ignore the bump at the end (as a couple of days ago I moved a few nodes over from a project that was out of work). Before that I was crunching very steadily. I also check regularly for frozen WUs (which aren't too common, but they will sit there forever as the watchdog is also frozen).

It might be a Windows problem then, if your Linux boxes have not dropped? I don't know. All I know is that my RAC is dropping like a rock.

Machines are all moblie XP2600's watercooled, running XP. And have been running these speeds for over 8 weeks. Bumped them up a little with the cool winter weather. No core will be allowed to run over 50c loaded.
1= 2700mhz was around 342ppd now 320
2= 2640mhz was around 337ppd now 310
3= 2596mhz was around 319ppd now 294
4= 2569mhz was around 332ppd now 308

RAC of each as I typed this.

Its not one machine, it's all of them equally.

They are hidden since one of my other posts when someone magically said what credit I was supposed to be getting, 1285 ppd comes to mind. But as stated my average was over 1330. Has been for over 8 months. So no reason for me to unhide them. Now its dropped under 1230, and still going down.

I have no wu's that have errored out for the last week. No machines are stuck. Just the PPD continues to drop like a rock.

I don't know if machines that are erroring out are dropping the rolling average. Or if the new Core Duos are having an effect, but my machines are turning out the same amount of work that they always have. They are just being granted less credit for the same amount of work done.

Something changed, and it's NOT ON MY END.

And don't say that one of the reasons given for the new credit system was not cross project parity. It was one of the biggest reasons given, along with optimized client over claiming.

And there is no cross project parity right now, not with these numbers.

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...

Hey, 3rd post ever, apologies in advance for being about as green as they come.
I've noticed that starting on Dec 22, that my WU's went from taking on average 10K+ CPU time to 85K CPU time. I though it was my rig(s) but I looked around at a few others crunchers computers (I apologize if looking at other peoples computers is in bad form) and have seen the same thing. Wouldn’t a drop in RAC be the result of ones computer simply taking longer to process these massive WU?
Or am I talking out of the back of the ship?
Happy Holidays all,
PCS

---

My drop in daily credits seems to have been from Dec 21-27. It appears that it has returned to about where it should be (i.e. 305/320 per day, instead of 77/320 per day).

I've noticed that starting on Dec 22, that my WU's went from taking on average 10K+ CPU time to 85K CPU time. I though it was my rig(s) but I looked around at a few others crunchers computers

A lot of you are not going to like what is planned for BOINC credit then ;-) but I'll just say it'll be quite a bugger if they implement it. You may get more credit if you are more useful to the project (i.e. have 8GB of RAM, or 4cores or somthing like that). cross project parity will be an teh project level no the individual computer level.


Cross project parity on the cpu level cannot be achieved anyway with the advent of optimised scinece applications so that should be ignored.
Since Rosetta is only loosely based on the benchmark, some computer are going to be higher than the average (boinc benchmark), e.g. Intels. AMDs seem to be even or lower.

As more intels P-M's and Core2's come on board they will cause the others to go down since they have to balance around that average some how.

Your RAC will not be constant and probably why the AthlonXP (mine had always claimed low compared to my other comps) just don't compute as fast as that average. Which is (in my thinking) why the RAC is dropping on them.
So as more of these faster PC's that crunch faster than the boinc benchmark indicates come online the rest should all drop unfortunatly.

That is my best unanted-explanation.

Though you are only under par compared to other projects if
in the case of, you claim <= granted
a) their projects give you too much credit
b) you use a optimised science app there, e.g. seti

OR
c) you regularly claiming less than granted.

---

Team mauisun.org

A brief on credits for those that are not familiar, then a few comments on the last several posts.

Credits are based on the work you crunch. If you have a quad core blade that runs fast, you will crunch a lot of work. You will claim a lot of credit (based on your stellar benchmarks) and you will be granted a lot of credit. You will crunch much more work per hour then your dual or single core counterparts, or machines with slower CPUs. Your credit claim and granted will reflect this.

The credit issued is based on a rolling average of credit claimed PER MODEL CRUNCHED. For WUs where models take a long time to run, the credit claim per model will be quite high, and so the rolling average will be high and the credit granted per model crunched will be high... commensurate with the additional time it takes to crunch that WU as compared to other types of WUs which complete models quickly.

@ paulcsteiner: You need to examine which location your machines are configured to (click on each machine on the participants page) and look at the bottom. Then go to your Rosetta Preferences for that location and set the work unit runtime preference to the level you desire. More info here.

If the erroring work units of other crunchers were being averaged in as zeros, or as the 20 credits I mentioned, we should see credit claims exceeding grants. Because the rolling average would be lower then it should otherwise be without the erroring WUs factored in. This is one of the reasons I asked about that.

If the average claims of other users are somehow noticeably reduced, that would cause such a problem. And if that were the case, we'd expect to see total credit for the project to be reduced as well (and it does seem to be the case), and you'd still see claimed credit near granted, but it would mean you must have previously seen more granted then claimed (or at least a higher ratio between the two).

The main way one might explain the claims of the average user declining would be if a significant portion of the user-base stopped using the optimized clients. I'm not up on them, has anything changed there that might cause people to have decided to stop using them? Another way would be if the standard BOINC benchmarks have changed in the recent releases. Does any way know if this has changed in a specific BOINC release?

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

My computers run with no graphics. I mention this because some people have had a lot of trouble with graphics recently.

I also have my crunch time set to 10 hours. Under some circumstances this can strongly affect a machine's RAC. Here's an example of why:

All models crunched for a particular type of WU will get the same credit. But suppose some WUs of that type take 1 hour per model, and others take 10 hours per model. Now let's say I download one of each, and so does someone else with identical hardware but a 1 hour crunch time. My computer with its 10 hour crunch time will crunch 1 model for the slower WU and 10 model for the faster WU, for a total of 11 models in 20 hours. The other person will spend 10 hours crunching 1 model of the slow WU (because Rosetta always finishes the model before comparing time used to time requested). The second WU will crunch 1 model in 1 hour, since this person is using a 1 hour crunch time. That's 2 models in 11 hours.

In this example I crunch 0.550 models/hour while the other person crunches 0.182 models/hour. Thus I end up getting 3 times the credit/hour as the other guy with the same hardware.

The above effect won't happen if the requested crunch time is greater than what is needed to crunch a slow model.

-_-

Nice, always one Axxxxxx.

To try start Sxxx.

"-_-" is starting trouble? I'm an "Axxxxxx"?

Well, "Fxxx" you. Me starting trouble would be to point out that you're acting like a 2 year old that soiled his diaper and blames everyone else for the inconvenience.

No problem with this at all. The problem I had was when a dedicated cruncher itself was earning 50%-60% less than it demonstratably capable of. Such a large variation should be an indication that something is wrong somewhere, either my cruncher itself or with the wu/project

A lot of you are not going to like what is planned for BOINC credit then ;-) but I'll just say it'll be quite a bugger if they implement it. You may get more credit if you are more useful to the project (i.e. have 8GB of RAM, or 4cores or somthing like that).

I don't know how many times I can type this out without banging my head against the wall any harder.

Lets try this one more time.....

I am actually running faster now (for about 8 weeks, ambient temps are lower allowing me to bump the speeds up a little) all 4 boxes are 100% stable with cpu core temps under 50c loaded.

The drop in credit started around the 6th of Dec., and has been going down ever since.

I am running the same times on my boxes. = not changed
I am running without graphics. = not changed
I am not getting any more errors. = not changed
I am running 24/7 on all boxes. = not changed
I am doing nothing different than before. = not changed
I have no viruses or any unknown processes running. = not changed

Nothing on my end has changed, only the credit awarded has changed. It's spread out equally over all 4 boxes. And is now about 100 PPD lower that it was for the last 8 months. Only in the last 2 weeks has something changed, and it's NOT ON MY END.

My teammate's RAC just below me has been dropping at just about the same rate as mine. He's running 2 AMD X2 dual core boxes. We both ran within 10-15 PPD of each other. He's down about the same amount as I am. So it's not just me.

I'm just searching for an answer... Does this coincide with the release of 5.43?

---

BETA = Bahhh

Way too many errors, killing both the credit & RAC.

And I still think the (New and Improved) credit system is not ready for prime time...