Hopefully you will see the graph which shows you what I am seeing

http://stats.free-dc.org/stats.php?page=user&proj=sah&name=71946

---

Has this been run up the flag pole to others ????

has what exactly been run up the flag pole? (to whom in particular?)

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

Richard
The reason for my original enquiry was that the reduction in average credit seemed to be on-going rather than sporadic. The occasions when the average goes up are rare. But as long as I can keep crunching, that's fine by me.

---

My RAC reminds me of Tom Petty's song Free Fallin

My user average chart looks like a waterfall :(

My RAC reminds me of Tom Petty's song Free Fallin

My user average chart looks like a waterfall :(

Dayumm, now while trying to Compare my ~20% credit drop at same angle range, to Richard's steady state, I have Tom Petty stuck in my head, ROFL

---

"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.

What I don't understand, at times I see my RAC leveling out, or a slight climb. Then look 2 hours later and it is back to a steady drop.

It's like there is something kicking in that lowers RAC :(

Hopefully you will see the graph which shows you what I am seeing

http://stats.free-dc.org/stats.php?page=user&proj=sah&name=71946

Don't see it there. I tried to post a graph but had the wrong one... I'll see if I can dig one one up, but yes, something has definitely changed.

Edit: The decrease can be seen here. Note the Credit per Day fall. (Ignore the Wednesdays as there is always a huge drop after the Tuesday outages.)

---

What I don't understand, at times I see my RAC leveling out, or a slight climb. Then look 2 hours later and it is back to a steady drop.

It's like there is something kicking in that lowers RAC :(

RAC is re-calculated every time there's a credit-related event, such as reporting completed work - I think whether the validator grants credit or not. Any previously earned credit is decayed (on a 7-day half life), and the new credit, if any, is added into the mix. That's one reason why we always advise people to measure performance on the reported runtimes of individual completed tasks, taking care to avoid overflows and match Angle Ranges.

The best reference on RAC is

http://web.archive.org/web/20120418125739/http://www.boinc-wiki.info/Recent_Average_Credit

My observation of non-decreasing RAC is based on three machines with nearly-identical hardware and similar operating regimes. They are:

host 3755243 - RAC 13,577.10
host 4292666 - RAC 11,631.04
host 7751088 - RAC 12,472.78

Similarities: Motherboard, CPU, RAM, disk drives, GTX 970, cases, fans, operating system are all fresh purchase in January this year to replace storm damage - I recycled the old host IDs.

Differences: 4292666 has a lower factory overclock on the GTX 750Ti
7751088 is used as a daily driver (I'm posting on it now), so has a core freed to handle all the browser tabs I keep open.

Operating parameters: SETI normally runs MB 24/7, but on the GTX 750 Ti only. Numberfields runs on the CPU, GPUGrid on the GTX 970, Einstein on the intel_gpu. Just occasionally 3755243 takes some time off to run Beta, Einstein got a NVidia boost when SETI crashed last month, and SETI got some 970 time when GPUGrid ran low on work. You can see that here (from 7751088):

Yes boincstats sure sows a decline, looking at my free-dc stats, they don't seem to match my BOING stats. Not sure why *puzzled*

What's particularly awesome, is that Richard's controlled thing is showing exactly what might be expected, While my own unounched and rarely looked at for a month Linux machine shows complete nonsensical behaviour.

How constantly are you riding that host Richard ?

---

"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.

Richard, how are you getting stats back from Dec 08? Both my computers only go back to March 22. But comparing your graph to mine, I'm see about double your decrease since March 22.

Looking closer at BoincTasks log, my spikes seem to match when I got a few AP tasks, but that is hard to tell, those tasks are long gone and hard to tell when they validated.

I've been watching this for a week or so, and there's no way it's an illusion. I started because I was trying to figure out why mine was dropping... ie machines not getting GPU work due to GUPPI? Technical issues ie bad cruncher spitting out Invalids? One of the ways it can be seen that RAC is constantly falling is to check the Top participants. This list updates infrequently, so is a snapshot of participant RAC at that point in time. Going into each profile shows their current RAC.

Here's what we see from the top 12 at the time I posted this:

Name		TP RAC		Current RAC	Change
Lee Klopp	714,808		714,145		-663
MJS		543,177		540,824		-2,353
BojanR		455,817		454,691		-1,126
WB5TUF - Glenn	449,755		448,701		-1,054
Igor_b		426,679		426,241		-438
Fawkesguy	381,914		381,269		-645
Mr. Kevvy	375,483		374,294		-1,189
msattler	308,949		308,329		-620
DARRENS		283,672		283,181		-491
Charles Long	270,878		271,526		+648
TSTC - Cloud...	270,009 	269,847		-162
toaster		260,086		259,899		-187

One increase and 11 decreases. Usually I don't even see the one increase, either. And this has been pretty much constant for over a week.

---

I have noticed a slight rise and fall in RAC, at the same time there has been a fall and rise in the numbers of validation pending.

---

Kevin

Ah yes, the natural beauty that is chaos manifests :-O

---

"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.

Yea it makes a difference with who you get paired with, I return everything in under 2 days,

But could get paired with a similar user, or 3 8 week time outs, it all makes a difference.

How constantly are you riding that host Richard ?

Far too much - I must get out more!

One thing may come as a surprise to readers - I run a 0.4 days (9.6 hours) cache on all three hosts - all on the 'store at least' setting, with zero additional. At the moment, I have 94, 60, 77 in progress on the three hosts I listed - they could each get 200 because the dual GPUs, but there's no need. Completing two tasks every 25 minutes at mid AR keeps them constantly churning, constantly requesting, and constantly topped up. If I do see a host running low (host 775 also runs BoincView showing 500 tasks across seven machines and four projects), I might poke it a bit, but usually there's no need, except perhaps at the end of maintenance. The trick - which on the whole BOINC manages by itself - is to find a set of compatible settings which keeps the cache level and doesn't allow anything to get backed off. My biggest bugbear at the moment is trying to snare three GPUGrid tasks, twice a day, to stop SETI spilling over onto the 970s. (Their PhD students have all graduated, and the new intake aren't producing research data yet.)

Richard, how are you getting stats back from Dec 08?

I used

<save_stats_days>N</save_stats_days>
How many days to save the per-project credit totals that are displayed in the Statistics tab of the BOINC Manager. Default is 30.

from Client Configuration Options.
But I had to use it before January 07 - once they're gone, they're gone.

Hmmm, now that I think about it...
Has anyone else noticed your tasks with an AR of around 0.42/0.40 are receiving much lower credit that the ~100+ they use to receive? I'm pretty sure that Angle Range use to net around 100 credits. Also, I'm pretty sure the GUPPIs were good for around 70+ credits when they first arrived, now they're in the 50s. Even the normal VLARs seem to be paying less, with most of them below 100.

I dunno, perhaps the GUPPIs are pulling down the CPU credit, creating a cascading downward spiral.
I suppose that would be the norm, the credit Always seems to fall when something new is introduced.
I could have sworn I use to receive around 100 credits for an AR 0.41, I still see a few but most are lower.

I've just done a quick check - six months ago 0.42is was giving between 95 and 105, compared to today's 75 to 80 - a twenty percent (round figure) reduction.....

Given the nature of both the algorithm and its implementation I am surprised the drop is only 20%!!!!


If one assumes that the reason for giving credit (new old or decrepit) is to be able to compare the performance of either one's computer, or the applications being run I think we have sufficient evidence to say that that the current system is of no use to man nor beast.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

For me the earlie Arecibo ~97-104 credit , 0.41 AR, seem to be around the 77-84 Cridt range, though clearly still settling down. Sadly the coming telescope soup is likey to make it much less simple, so I've pretty much resigned myself into buying a VW combivan and touring the world.

---

"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.