My i7 CPU had an RAC decline beginning in early December and another CPU in late December and the other 2 had declines after January 1. Each shows host drops so it is not 1 computer. Before December my approximate RAC was about 67K and now it it about 62K or a drop of 5K with no sign of recovery. Has anyone else seen similar behavior?

---

Delta-V

First, If you do astropulse, availability has been more limited, as there seems to be a rush on those when released due to higher credit awards.

Since then, some of us have worked on unravelling some of the mysteries.

In simplest terms, the CreditNew mechanism is 'chaotic', having 4 known design limitations/flaws currently under study. 2 of these would be current and may be applicable to what you see. another 1 (of 4) will arise with future applications, and the last connected to work time estimates.

In general, with similar actual throughput, you should see some slight boom or bust cycles (+/-20%), not unexpected, and 'self similar' patterns over medium to long term.

If you examine the project supplied original minimal work estimates (# operations required to perform the task as a minimum), and compute credit against the cobblestone scale, you'll see that credit for multibeam comes to less than 1/3.3, and AP less than 1/1.6 expected.

SO basically the current mechanism 'believes' in magic, by ability to compute tasks in less than the amount dictated by the laws of physics.

It's broken and not a useful measure of squat.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Jason you have posted some concise ideas on CredNew recently, have they been dicussed with Dr. Anderson? I sure hope so.

---

Jason you have posted some concise ideas on CredNew recently, have they been dicussed with Dr. Anderson? I sure hope so.

Certainly the more diplomatic members of the team have opened avenues there. I'm from a mixed computer science and Engineering background, so am less diplomatic and tend to steer toward the position of 'Your stuff's broken, you better fix it', While David would apparently prefer finished/tested patches rather than looking at it himself. Fair enough. All it really means is he doesn't understand his own system and needs serious help. Let's hope I can find the resources I need to provide that help.

[Edit:] Bernd from Albert@home has already stepped forward with some of the beta resources needed, though they're suffering their own issues at the moment. It'll be a fairly long involved process.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

My i7 CPU had an RAC decline beginning in early December and another CPU in late December and the other 2 had declines after January 1. Each shows host drops so it is not 1 computer. Before December my approximate RAC was about 67K and now it it about 62K or a drop of 5K with no sign of recovery. Has anyone else seen similar behavior?

I only crunch MB, and RAC has been falling for some time.
There was a big decline starting around the beginning of Dec, and system outages added to that fall. Late Dec/early Jan after the outages RAC came back up, but still not back up to the early Dec level (which was still less than it was after the initial plunge with the introduction of v7). By the end of Jan it had reached a new high, which of course was lower than any previous level. After that it dropped again to a new low high level.
Since then we've had several system outages (pretty much one every week, some for a day, some for a couple of days) & that has caused RAC to drop significantly.
Since then I've replaced my existing video card with 2 new ones so any new high levels of RAC will be higher than they otherwise would have been.
So far, with almost double the processing being done, I've managed to get back to the late Jan level, which is around 1,100 less than it was after RAC stabilised after v7 was introduced.
In otherwords I've had to almost double my processing power to get just half the credit I was prior to v7.

Credit New isn't broken, it's completely screwed.

---

Grant
Darwin NT

Credit New isn't broken, it's completely screwed.

I actually went in with that expectation, seeing the behaviour as an end user like everyone else. In the past I've had little interest in Boinc code other than to make my own work...work.

Where it became interesting is that logically there are no code 'bugs' found (yet), and many of the base design decisions appear sound. There are just some major glaring design omissions with respect to stability and control input assumptions. Those particular omissions are more than enough to produce the behaviour we see (and worse).

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

+1

+ another one: How is Your "Task duration correction factor" ????

Mine has been one for about a week or so.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

+1

+ another one: How is Your "Task duration correction factor" ????

Mine has been one for about a week or so.

Modern Boincs have it locked to one (at projects that don't use DCF), the APR for different app versions changes instead.

Claggy

+1

+ another one: How is Your "Task duration correction factor" ????

Mine has been one for about a week or so.

With relatively recent Boinc client and stable unchanging system and usage poatterns, it should always be around one, as this project locks project DCF, and CreditNew adjusts the estinates server side.

For myself though, I've stuck to a customised old 6.10.58, where I've uncoupled project DCF from the estimates and injected per application DCF, PID controlled, instead. That pretty much allows the estimates to track with hardware change, and quickly converge during machine usage variation. I've yet to find any particular reason for removal of local client estimate tracking, apart from that Project DCF didn't cope with multiple applications per project, or hardware change

[Edit:] Oh, and that Project DCF was a bit unstable. In fact the Server side tracking for CreditNew is two cascaded DCFs, and has similar instabilities exacerbated by the time between validations.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Credit New isn't broken, it's completely screwed.

Because that the right name of this Random Number Credit Generator must be: CreditScrew!

---

+1

+ another one: How is Your "Task duration correction factor" ????

Mine has been one for about a week or so.

Computer 6896459: Task duration correction factor; 1 - GTX670 i7 960 3.2GHZ Win 7 Pro x64
Computer 6740220: Task duration correction factor; 8.786986 - GTX470 Core2 Quad Win XP Pro x64
Computer 6024182: Task duration correction factor; 1 - GTX560Ti P4 3.6 GHZ LGA 775 Win XP Pro x64
Computer 6650722: Task duration correction factor; 8.700757 - GTX470 P4 3.6 GHZ LGA 775 Win XP Pro x64

Should the GTX470s have such odd correction factors?

---

Delta-V

+1

+ another one: How is Your "Task duration correction factor" ????

Mine has been one for about a week or so.

Computer 6896459: Task duration correction factor; 1 - GTX670 i7 960 3.2GHZ Win 7 Pro x64
Computer 6740220: Task duration correction factor; 8.786986 - GTX470 Core2 Quad Win XP Pro x64
Computer 6024182: Task duration correction factor; 1 - GTX560Ti P4 3.6 GHZ LGA 775 Win XP Pro x64
Computer 6650722: Task duration correction factor; 8.700757 - GTX470 P4 3.6 GHZ LGA 775 Win XP Pro x64

Should the GTX470s have such odd correction factors?

You're running a relatively old Boinc version (now), while Boinc 7.0.64 doesn't use DCF for the Seti project, it also doesn't set it to One like later Boincs do, so it remains at what it was.

Claggy

My i7 CPU had an RAC decline beginning in early December and another CPU in late December and the other 2 had declines after January 1. Each shows host drops so it is not 1 computer. Before December my approximate RAC was about 67K and now it it about 62K or a drop of 5K with no sign of recovery. Has anyone else seen similar behavior?

Looking at one of my machines that only runs MB work:


It looks more like there was a rise in November and things are pretty consistently bounding around at the moment. Having large amount of VLAR or VHAR tasks will also tend to effect the RAC.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Cool trick! How did you get BM to show more than a month in the graphs?

Cool trick! How did you get BM to show more than a month in the graphs?

Use <save_stats_days>365</save_stats_days> to show a year:

Client configuration

Claggy