I have a q6600 running at about 3.1 ghz and doing 100 percent AP. RAC is now about 8800 pending credits are about 9:1

Recent average credit 167,088.37

Pending credit: 1,314,126.57

7.86:1

Still hanging out near 8:1

---

--
Classic 82353 WU / 400979 h

Yup, the RAC is starting to drop again and the pending is going up. Looks like I've picked up about another 12000 pending over the weekend...

Pending credit: 2,762,176.75

8.7:1

---

Mine's not that spectacular, but..

pending: 54,533.26
RAC: 5,656.20

9.64:1

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

I'd like to propose a theory and a question:

I'm wondering if the new BOINC is causing some of the pending credit growth and temporary RAC decline?

To explain:

Old BOINC processed workunits in the 'natural' order they arrived. FIFO, except when some shorties came along and needed priority. Otherwise, FIFO.

New BOINC processes workunits in strict 'Report deadline' order, and this is not necessarily FIFO. This is a big change.

How does this affect RAC and pending credit? Well, if the majority of users are still on BOINC 6.4.x, and you are using 6.6.2x, then your processing is time-displaced from the processing of your wingmen. This will sort itself out in a short time, your RAC will eventually recover, but your pending credits will remain higher than in the past.

Example: You are using BOINC 6.6.20. Your wingman is using BOINC 6.4.7.

The following numbers WU's come into both of your worlds:

1,2,3,4,5,6,7,8,9

Wingman processes the WU's in this order:

Wingman, FIFO order: 1,2,3,4,5,6,7,8,9

If you did not switch to 6.6.20, you would process in the same order. However, you are now processing in a different order. I'll use the worst case example to test my argument:

You, Report Deadline order: 9,8,7,6,5,4,3,2,1

You see, WU number 9 is completed by you immediately, but will sit in your Pending Credit bucket until the wingman finally processes it, maybe 5 or 10 days from now! Also, your wingman will see an increase in pending credit and a temporary decrease in RAC since he or she is waiting for you to finish WU number 1.

RAC will eventually recover, but at a total loss over time, relative to the increase in everyone's pending credit, at least until every single user converts to the new non-FIFO processing order by upgrading their version of BOINC.

And that is why RAC is dropping or barely growing right now, and pending credits are on a big upswing. Thoughts?

Bob Mahoney

---

Opinion stated as fact? Who, me?

From 6.6.23 the FIFO order for CUDA units is implemented. So however much of an effect this has, it will be temporary.

F.

---

...
And that is why RAC is dropping or barely growing right now, and pending credits are on a big upswing. Thoughts?

Bob Mahoney

The default settings for "Connect interval" and "Extra work" are 0.1 and 0.25 days. I have a feeling that maybe half the users have never modified their preferences nor felt an obligation to install the latest post-alpha version of BOINC. Perhaps that's also the half running slower hosts and/or turning them off at night, though. Anyhow, hosts running in that fashion provide some buffer, they have both short queues and FIFO processing so have little effect on wingmates' pending.

OTOH, every time there's a server problem affecting work delivery that motivates more users to set long queues. The average turnaround time for Enhanced work may have grown somewhat though it is so affected by the current work mix that's hard to tell.

It would be interesting to compare the RAC curve of a strong host not running CUDA to yours with. That would be the best way to estimate the amount of effect using deadline order for CUDA is having. I know that _heinz non-CUDA 8 core system has also had a recent RAC decline and pending growth, for instance.

                                                                Joe

Pending credit: 49,248.49 [incl. CUDA overclaim! AFAIK: - ~ 26 % really]

36,443.88 pendings..


RAC 14,469.19


Ratio: 2.518722886353693606898520235065 ;-D


Current the RAC isn't stable.. because no stable tasks/cache..
Normally it was ~ 15,500 - 16,000 RAC

Current pending credit: 40,345.50
Subtraction CUDA overclaim: 29,855.67 [AFAIK: - ~ 26 %]

RAC 15,455.90

This would mean a ratio of: 1.9316681655549013645274620048008


The average turnaround time of the results are nearly similar to my tasks cache settings.
A well sign for stable RAC and stable pending credits.

---

...
CUDA overclaim: 29,855.67 [AFAIK: - ~ 26 %]
...

I did some checking awhile ago, it seems to range from ~ 89% to -2%(underclaim) depending on AR. The 26% approximation seems fair, though.

                                                                Joe

Bob, et al

The Server Status page tells alot. For a couple years since Matt implemented that section about

Result turnaround time (last hour average) 77.03 hours 168.97 hours 0m

It shows 3 days Average turnaround for MB and 7 days for Astropulse

From My own observations, I know that when the validators are clogged pending Credits rise fairly quickly. If you add stuck databases and the validator clogg it gets worse faster.

So While I do not know the actual percentage of users that have 10 Cache/buffers. It does appear that the 90+ percentile return workunits in 3 days. If I compare my RAC to my Pending Credits it is close to 3 days worth of work. The exception is Astropulse (but that is on a faster machine, which holds close to the 7 days). Mathmatically even that works out +or- within a percentage.

Hopefully Tues Server maintainence will reduce some of the latest Clogg in the pipes...

Regards

I'd like to propose a theory and a question:

I'm wondering if the new BOINC is causing some of the pending credit growth and temporary RAC decline?

To explain:

Old BOINC processed workunits in the 'natural' order they arrived. FIFO, except when some shorties came along and needed priority. Otherwise, FIFO.

New BOINC processes workunits in strict 'Report deadline' order, and this is not necessarily FIFO. This is a big change.

How does this affect RAC and pending credit?

Bob Mahoney

---

Please consider a Donation to the Seti Project.

It will be interesting if someone can generate a histogram for results turn around time. Is the data available in some xml file somewhere that all the sites used to generate the data?

The other interesting metric is the histogram for WU cache of host population. This I suppose will be harder to generate.

Without hard data it will be hard to try to statistically quantify the effects contributing to the long pending credits and turn around time.

Personally I'm on a 9 or 10 days WU cache, and my hosts sometimes do the strange thing of downloading shedloads of APs, run them until its about clean in 15+ days, so in effect will return results in a uniform distribution from times of 1-2 days to 15 day or so... apologies to wingmen of course, that you can only get validation with a 7 days mean if you caught one of mine.

Edit: BTW this thread is getting a little long in the tooth so perhaps the owner will consider turning it into a persistent thread with a new revision soon (e.g. Milestones or Panic threads)

---

The turnaround time for my mostly AP cruncher is 2.78 Days.

So to the other 90%+ out there doing AP slowly -

Lift your game, you're holding my RAC back . . . . LOL

On a very un-scientific survey of my Q6600 that only does AP, unless there are none available, shows that about 2/3's of them are validated within 6 days.
It has a 2.5 day cache.

It does about 6 tasks/day, and 9 tasks were completed over a month ago.

My RAC:pending is ~10:1

My equally unscientific survey has me with a turn around time of 5.6 days, so about the same. I think that we are holding Virtual Boss back so we need to shovel more coal, step up our game, give 110% and other sports clichés. LOL.

Present RAC : pending is 7 : 1

My ratio has come down a lot. I guess it was just a blip caused by starting AP and using optimized apps.

---

Pure mathematics is, in its way, the poetry of logical ideas.

Albert Einstein

Thanks kenzieB, my RAC jumped +150 almost as soon as you posted . . . LOL

It will be interesting if someone can generate a histogram for results turn around time. Is the data available in some xml file somewhere that all the sites used to generate the data?
...

Scarecrow gathers and graphs the Server Status numbers, the 90 day version for Astropulse or S@H Enhanced gives the longest available view. I don't know of another source for historical data of tha kind.

                                                               Joe

I have some AP WU's in the Pending McGuffin from mid March.
A good proportion of my outstanding 90+ are with wingmen with no
chance of completing in alloted time.
There should be some kind of mechanism that only sends AP's to
proven AP crunchers. I.E. send 1 back in good time get another
send a second back in good time get 3 and build up.
Or fail to send back in good time reduce the ammount
of AP's sent to the host.

---

I thought my pending stabilized and that my RAC would now steadily increase.. but no..:
Pending:111,966.08 (got +10 within the week)
RAC : 12,785 (about stable the last few days..)

Ratio 8.8 : 1

---

The SETI@Home Gauntlet 2012 april 16 - 30| info / chat | STATS

Ah I'll see what keeps pending :)

My RAC is still climbing broke 3000 now and I have a pending amount of 42k+
so

I'd say I'm at 14:1 (If I crunched it ok.)

I'm using only astropulse v5.03 (SSE3 optimized by AK)
Turnaround is about 20-22hrs or so on my 1st system
Turnaround is about 52-56hrs or so on my 2nd system

my 1st system is a 2CPU QuadCore Opteron64 2.3 Ghz (2376 Shanghai) 8 threads
my 2nd system is a 1CPU Single Athlon64 3000+ 2.0Ghz, 1 thread

I expect to get to 9k RAC, maybe even 10k I'll see. I expect pending credit to normalize then.

---