I'm fairly sure BS uses a different formula to calculate RAC (but not 100%)...

Even so, their formula would have to based upon data pulled from Rosetta and other projects on what is usually a daily window. So there is still a timing issue as to exactly when a client machine reports completed tasks, and when fresh data is exported each day.

---

Rosetta Moderator: Mod.Sense

This machine reports every 2-4 tasks completed. Whenever Boinc mgr calls for it.
I have it all set to default other than the amount of extra work stored and that it should be computing at all times on all cores at 100%.

BS seems to compile there RAC graph data every 2 days. Last compilation was the 29th. Their text compilation of RAC and other stats was yesteday. They showed at that time I had 1392 RAC which looks about right. Unfortunately I had a windows error that started dropping my cpu usage to 33% on one core and 20 something percent on the other. (at the minimum that is 4 work units that were run at less than full power). I restarted and then ran into a task failure. So my RAC for the past 48 hrs has been dropping. I will keep an eye on it and hope these t448 tasks don't crash to much. Then I will compare BS to RAH and see how they match up.

I'm fairly sure BS uses a different formula to calculate RAC (but not 100%)...

Even so, their formula would have to based upon data pulled from Rosetta and other projects on what is usually a daily window. So there is still a timing issue as to exactly when a client machine reports completed tasks, and when fresh data is exported each day.

Thank you for crunching R@H. It is a great project.

RAC does move up and down a little. It can even change on the same machine from week to week based on the work units and other factors.

You might get a better picture if you look at your weekly totals. RAC moving a few points day to day is no big deal. If you see a steady change, you may want to investigate. 20 points over a few days is not a big change at all.

RAC is also based only on credit granted. Credit claimed has nothing to do with the calculation. Even if your credit granted exceeds credit claimed on every work unit, your RAC could go down.

Good luck crunching.

---

Thx!

Paul

the credit claimed is not what i was talking about, my credit granted has for 99% of the time exceeded claimed credit anywhere from 3 points to like the latest work unit as of the time of this post where claimed was 93.75 and the granted was 122.88 and so on. I have had one failed task and one task that actually granted below claimed by a few fractions of a hundredth of a point and I had a windows problem that dropped my cpu usuage for a day to 30% on one core and 25% on the other and I know that caused my RAC to drop along with the other problems.
But my point is and has been that if 97-99% of the time this machine keeps that kind of average up of getting more points granted than claimed then it would make sense that the RAC graph would continue to climb until something really drastic such as going offline or the above problems. Now that those issues have been resolved I will watch the RAC and see if it continues to climb.

Thank you for crunching R@H. It is a great project.

RAC does move up and down a little. It can even change on the same machine from week to week based on the work units and other factors.

You might get a better picture if you look at your weekly totals. RAC moving a few points day to day is no big deal. If you see a steady change, you may want to investigate. 20 points over a few days is not a big change at all.

RAC is also based only on credit granted. Credit claimed has nothing to do with the calculation. Even if your credit granted exceeds credit claimed on every work unit, your RAC could go down.

Good luck crunching.

the credit claimed is not what i was talking about, my credit granted has for 99% of the time exceeded claimed credit anywhere from 3 points to like the latest work unit as of the time of this post where claimed was 93.75 and the granted was 122.88 and so on. I have had one failed task and one task that actually granted below claimed by a few fractions of a hundredth of a point and I had a windows problem that dropped my cpu usuage for a day to 30% on one core and 25% on the other and I know that caused my RAC to drop along with the other problems.
But my point is and has been that if 97-99% of the time this machine keeps that kind of average up of getting more points granted than claimed then it would make sense that the RAC graph would continue to climb until something really drastic such as going offline or the above problems. Now that those issues have been resolved I will watch the RAC and see if it continues to climb.

Your claimed credit has no effect on your RAC, so even if your claimed credit was 10 on each task, and your granted was 100, your RAC could be falling. ALso, even if you had a computer that ran R@H 24/7 perfectly the RAC would vary because you might get a task that takes 1hr per decoy, while an identical machine may get a task from the same family which happens to take 56 mins to do a decoy. Both will get the same credit, although they haven't done quite the same amount of processing...

HTH
Danny

---

the credit claimed is not what i was talking about, my credit granted has for 99% of the time exceeded claimed credit anywhere from 3 points to like the latest work unit as of the time of this post where claimed was 93.75 and the granted was 122.88 and so on. I have had one failed task and one task that actually granted below claimed by a few fractions of a hundredth of a point and I had a windows problem that dropped my cpu usuage for a day to 30% on one core and 25% on the other and I know that caused my RAC to drop along with the other problems.
But my point is and has been that if 97-99% of the time this machine keeps that kind of average up of getting more points granted than claimed then it would make sense that the RAC graph would continue to climb until something really drastic such as going offline or the above problems. Now that those issues have been resolved I will watch the RAC and see if it continues to climb.

Your claimed credit has no effect on your RAC, so even if your claimed credit was 10 on each task, and your granted was 100, your RAC could be falling. ALso, even if you had a computer that ran R@H 24/7 perfectly the RAC would vary because you might get a task that takes 1hr per decoy, while an identical machine may get a task from the same family which happens to take 56 mins to do a decoy. Both will get the same credit, although they haven't done quite the same amount of processing...

HTH
Danny

now wait a minute...your telling me its based on numbers of decoys produced?
i guess i have to back and read that link that mod put in here a long time ago...none of this makes sense now. first i thought it was based on granted credit, then i get this story that it could fall due to computing differences?
i am now totally confused.

AFAIK, BS-RAC is based on a 30-day moving average of daily credit. They do provide a separate RAC figure based on exponential decay that BOINC uses, though. Its not shown on the signatures but is provided on the detailed stats page.

---

now wait a minute...your telling me its based on numbers of decoys produced?
i guess i have to back and read that link that mod put in here a long time ago...none of this makes sense now. first i thought it was based on granted credit, then i get this story that it could fall due to computing differences?
i am now totally confused.

I'm saying that your RAC is based on your granted credit, which is based on the number of decoys produced * value of those decoys. If you crunch 10 decoys in an hour on a particular workunit, there's a chance that another identical machine gets lucky and crunches 10 decoys in slightly less time. In that case, the second machine will get slightly more credit per CPU second. This is all regardless of the claimed credit score which is based on your BOINC benchmark.

---

greg, doesn't it make sense that if you have had a few little things occur that you feel may have taken you down to 97% of theorectical capacity, that your RAC should drop about 2 or 3%? It actually seems to validate that RAC very accurately reflects how much work your machine is producing.

And yes, granted credit is always based on models produced. Models are the useful unit of work to the science. That is why granted and claimed are listed seperately. And then RAC is the average over time of the granted credit.

---

Rosetta Moderator: Mod.Sense

now wait a minute...your telling me its based on numbers of decoys produced?
i guess i have to back and read that link that mod put in here a long time ago...none of this makes sense now. first i thought it was based on granted credit, then i get this story that it could fall due to computing differences?
i am now totally confused.

I'm saying that your RAC is based on your granted credit, which is based on the number of decoys produced * value of those decoys. If you crunch 10 decoys in an hour on a particular workunit, there's a chance that another identical machine gets lucky and crunches 10 decoys in slightly less time. In that case, the second machine will get slightly more credit per CPU second. This is all regardless of the claimed credit score which is based on your BOINC benchmark.

that idea is stupid if you ask me...RAC for ME should be based on MY system crunching and not my system vs another system. Also as far as I know each system gets one unique job unless it fails on another system. so how can you compare one unique job vs another unique job? no two jobs start at the same random point, so again how can you compare? also you can not compare two machines as they may not be set identical. I OC my machine to a certain level and have so much memory and run 24/7. What machine is going to have the same identical setup and OC speed and also run 24/7 and get the same tasks I do?

again...how does a GRANTED credit level that is always above a claimed credit level 99.95% of the time equal a DROP in RAC instead of climbing. even based on 30 days the GRANTED credit runs higher by a certain level over claimed credit.
looking back to my oldest task that is still listed, the claimed credit was in the 84-96 range and the GRANTED credit runs from a low of 77 to a high of 161.
There was 1 no report error 28 july (no credit asked or given), 1 crash (26 july) , two problems with windows (26 july) that caused my system to run at less than max and one full run at OC that somehow came in at .41 under the granted. After that everything runs in the range stated with the latest numbers showing 92 claimed vs 110 granted. So I don't understand how I just started to gain my RAC back and then less than 24hrs later I drop again. The numbers just don't add up. So lets see...7 days back from today is the 27th.

Here are the totals from the oldest task still listed to the most recent

27th
94.79 100.02 - earliest task
94.12 114.49
91.29 110.06
92.42 129.04
91.94 105.55
91.90 129.71

28th
93.54 125.05
93.50 116.20
92.61 111.47
91.86 101.76
93.41 114.09
91.44 108.31
94.94 111.74
94.00 111.10
92.07 123.55
95.25 108.97
88.25 111.70
90.69 114.47

29th
92.91 89.03
91.95 126.77
94.36 132.21
92.05 116.77
91.35 107.65
92.11 111.17
93.27 119.00
92.39 116.29
97.38 84.22
94.44 123.91
88.95 102.14
93.91 133.46

30th
93.38 132.22
89.44 102.62
93.57 127.78
94.77 112.80
92.58 134.27
94.10 128.59
93.59 106.60
94.96 110.00
94.92 111.23
94.22 115.86
94.77 115.85
93.55 109.85

31st
93.89 110.17
91.17 96.92
90.83 80.19
94.79 97.35
93.94 106.63
104.82 122.63
20.02 ---
88.40 84.17
84.35 77.90
93.74 117.89
93.52 101.48

Aug 1
96.62 96.21
94.42 161.34
89.24 135.32
95.10 108.14
95.25 105.63
94.97 109.55
94.74 117.05
95.46 130.98
94.64 110.97
94.68 113.02
94.03 111.53
94.26 109.49


Aug 2
96.70 86.16
94.09 110.95
93.08 131.11
94.46 117.49
90.98 126.89
95.32 112.54
94.72 125.68
93.04 119.77
93.82 119.54
93.75 122.88
95.27 130.90
95.59 117.78
94.90 125.17

Aug 3
93.11 124.92
102.79 112.36
84.37 93.63
86.18 93.63
92.70 124.71
92.40 110.48
92.69 108.52
92.63 112.40
92.45 128.47 - latest task reported (and the RAC graph shows downward after a brief climb earlier today and yesterday)


As said earlier..I can not see how in the world I should be taking a drop on MY RAC when I have totals like these. Even with defective work units and under claimed credit totals on perhaps 3 or 4 tasks the others make up for that. Like a 92.70 claimed and 127.71 granted giving me a 35.01 point gain versus one day earlier with 96.70 claimed and 86.16 granted which is a 10.54 loss. The difference is 24.47 gain still in just 24 hrs. I should not even see a drop for that.

You have struck upon the very reason that credit is based on work done. But since all proteins are different, you can just say each model is worth 20 points (or any set number). So, instead, to automatically adjust for the relative difficulty of each type of protein and each method of studying it, a rolling average of credit claims reported by others for other models on the same piece of work are used to determine credit per model issued.

---

Rosetta Moderator: Mod.Sense

From the beginning:

Claimed credit is your BOINC benchmark * time taken on the task.

The BOINC benchmark is a very simplistic (Whetstone + Dhrystone) / 2 and so doesn't take into account cache or memory speed etc.

Granted credit is the number of decoys your computer produced * the value of those decoys.

The value of the decoys is the average of the claimed credit (per decoy) from all previous submissions of the same type of task.

For example, there is a Work Unit released called Protein1. The first computer to report that task (probably a bakerlab in-house computer) gets whatever it claims for that task, which will be its BOINC benchmark * time taken. Say that results in it getting 4.1 credits per decoy. The next computer that reports that task gets the average of 4.1 and its claimed credit. Say that computer had a slightly higher benchmark and so the average was increased to 4.2 per decoy. If your computer is the 50th to report a task from this Work Unit then its benchmark will be of little significance as it will have little effect on the average, and therefore on the granted credit.

The system relies on the fact that decoys within a given Work Unit will generally require a similar amount of processing to complete. There is variability within them, but it tends to average out, even within a task if it contains more than a couple of decoys.

Using your situation as an example now, lets say your benchmark is low compared to the CPUs actual throughput (which would be correct for a Core2/Phenom CPU, and even more so for a Penryn, as its real-world throughput is much better than an older CPU with the same benchmark). Your claimed credit is therefore low. When your completed task is submitted, your claimed credit will be lower than your granted credit because your CPU produced more decoys than the benchmark would suggest.

However, if you run a task and get lucky - i.e. the decoys take less processing that the average for that Work Unit - then your granted credit will be high. If you then get a task of the same type and aren't so lucky - i.e. the decoys happen to take a lot of processing - your CPU will produce fewer decoys in the same time, and will therefore get less credit when they're reported.

Throughout this your claimed credit will be less than your granted credit, but your RAC will fall when the 2nd task is reported as you're getting less credit per second.

That's just an example, but hope it clarifies why claimed and granted credit aren't really linked, and that your computer isn't being compared to any others except purely in terms of processing power. There is some natural variability but it's fairly minimal most of the time, and affects everyone equally.

HTH
Danny

---

ok thanks for the explanation.

the only thing i still don't follow is that my granted credit has always exceeded the claimed credit except for the few cases were it bogged down.

Based on the totals below which you guys still have not addressed it would say that my RAC should be going up since it consistently the case (with a few exceptions) that granted is higher than claimed.

Look at the last task I put on here. 92.45 claimed 128.47 granted yet my RAC graph shows a downward trend. Look at the work before that: 92.63 112.40 and the day before the last 3 tasks: 95.27 130.90 95.59 117.78 94.90 125.17. Again all positive point gain.

Now tell me that with those kind of results...is my RAC supposed to be up or down?
If we were to go back to a week before this last week, the results would be the same.

Based on this wiki paragraph: ...it tracks the rate at which you are earning credit with Credit earned today having more weight in the average than the Credit granted in the past; each day farther back the less impact on the "running" average.

Again I would say that my RAC should be climbing or holding steady and not dropping.

Another thing that did not happen that would cause a drop or a flat line would be this: So, if one or more of your machines stops processing work for a BOINC Powered Project, it is possible that its Recent Average Credit will remain at a constant value for some period of time and you will not see an expected decay curve.
But that has never happened. It has been on 24/7 for over a month now.
So this example is out.

So again..the logic says and my rough understanding of the math in simple terms..is that as long as the granted credit is consistently high than claimed credit I should have a upward line over a weeks time. I have shown you a weeks worth of work totals and as you see only a few went below claimed. But the next task goes real high making up for the low mark and still giving me a positive claimed credit.

I still do not understand...and please explain it to me as simply as you did before...why after a week of higher than claimed credit results in the granted credit..why my machine is showing a downward graph instead of flat line or upward?

Based on the totals below which you guys still have not addressed it would say that my RAC should be going up since it consistently the case (with a few exceptions) that granted is higher than claimed.

Look at the last task I put on here. 92.45 claimed 128.47 granted yet my RAC graph shows a downward trend. Look at the work before that: 92.63 112.40 and the day before the last 3 tasks: 95.27 130.90 95.59 117.78 94.90 125.17. Again all positive point gain.

This is where your logic is failing - granted being higher than claimed is irrelevant for your RAC.

---

Based on the totals below which you guys still have not addressed it would say that my RAC should be going up since it consistently the case (with a few exceptions) that granted is higher than claimed.

Look at the last task I put on here. 92.45 claimed 128.47 granted yet my RAC graph shows a downward trend. Look at the work before that: 92.63 112.40 and the day before the last 3 tasks: 95.27 130.90 95.59 117.78 94.90 125.17. Again all positive point gain.

This is where your logic is failing - granted being higher than claimed is irrelevant for your RAC.

ok...i'll take that.
Then what totals or combination of totals increases or decreases RAC?
This is where I am having a hard time understanding.
Boinc wiki talks about both equaling total credit and then doing some sort of formula which I don't understand to get RAC. Can you explain that simply? Thanks

ok...i'll take that.
Then what totals or combination of totals increases or decreases RAC?
This is where I am having a hard time understanding.
Boinc wiki talks about both equaling total credit and then doing some sort of formula which I don't understand to get RAC. Can you explain that simply? Thanks

I've cut from my previous post, and tried to tidy it up a bit:
Using your situation as an example now, lets say your benchmark is low compared to the CPUs actual throughput. Your claimed credit is therefore low. When your completed task is submitted, your claimed credit will be lower than your granted credit because your CPU produced more decoys than the benchmark would suggest.

Lets say your AVERAGE CLAIMED credit is 50 credits per hour from your benchmark, which we've established is lower than it should be for your CPU.
Lets say your AVERAGE GRANTED credit is 70.

If you run a task and get lucky - i.e. the decoys take less processing that the average for that Work Unit - then your granted credit will be high - lets say AVERAGE GRANTED credit goes up to 75.

If you then get a task of the same type and aren't so lucky - i.e. the decoys happen to take a lot of processing - your CPU will produce fewer decoys in the same time, and will therefore get less credit when they're reported - lets say AVERAGE GRANTED credit falls to 65.

Your RAC will fall when the 2nd task is reported as you're getting less credit per second (Granted RAC falls from 75 to 65). Throughout this your granted credit is higher than your claimed credit (Average of 50), but your RAC is still falling.

That make sense?

Danny

---

ok...i'll take that.
Then what totals or combination of totals increases or decreases RAC?
This is where I am having a hard time understanding.
Boinc wiki talks about both equaling total credit and then doing some sort of formula which I don't understand to get RAC. Can you explain that simply? Thanks

I've cut from my previous post, and tried to tidy it up a bit:
Using your situation as an example now, lets say your benchmark is low compared to the CPUs actual throughput. Your claimed credit is therefore low. When your completed task is submitted, your claimed credit will be lower than your granted credit because your CPU produced more decoys than the benchmark would suggest.

Lets say your AVERAGE CLAIMED credit is 50 credits per hour from your benchmark, which we've established is lower than it should be for your CPU.
Lets say your AVERAGE GRANTED credit is 70.


Let me chew on that for awhile....being I am at work I can't focus on this.
I will also follow this with my current tasks.

thanks for the help so far..greg
If you run a task and get lucky - i.e. the decoys take less processing that the average for that Work Unit - then your granted credit will be high - lets say AVERAGE GRANTED credit goes up to 75.

If you then get a task of the same type and aren't so lucky - i.e. the decoys happen to take a lot of processing - your CPU will produce fewer decoys in the same time, and will therefore get less credit when they're reported - lets say AVERAGE GRANTED credit falls to 65.

Your RAC will fall when the 2nd task is reported as you're getting less credit per second (Granted RAC falls from 75 to 65). Throughout this your granted credit is higher than your claimed credit (Average of 50), but your RAC is still falling.

That make sense?

Danny