Got a call from some guy in Zimbabwe this morning.
Said he could get me a Seti toaster for my Seti credits.
Apparently the exchange rate is rather good there.

MeowLOL.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Can you order me one as well please Mark - you'l have to send your bank details to my Nigerian banker so we can complete the transaction...

---

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

Yeah, that was me but via Pm where I asked you if the number was lower or not.

And as far as I can tell, I got all the "synthetic benchmark" replies for nothing. I was right all along.

When I said "you can't go beyond 100% efficiency".... well that's the number Eric is quoting "very, very carefully".

But before we go into that, I've been slow-smoking a 4kg slab of local-farm pork ribs on the Webber for the past 4hrs... with Apricot & Pomegranate wood chunks I cut off from the yard trees last year. Trees that never saw fertilizer or pesticide in their over 20 year life span. Daughter has been running around at peek-cuteness using single words to convey her thoughts while I keep smiling, pretending there's no chance she's gonna crack her head open any second. Oh and lazily downing three pints and munching on this morning's catch-of-the-day fish (prepared by Daphne's very Greek and very Great-Grandma). Fish that's actually tasty since it isn't from an ocean :) I also see my chili pepper plants are coming along beautifully too... Damn, that Apricot smoke smells insanely good.

But now time for a quick siesta because in three hours the ribs will be ready and I need to wake up before then to prepare the salads and some other stuff. And more beer (obviously).

I'm getting old. What were we talking about again?

Can you order me one as well please Mark - you'l have to send your bank details to my Nigerian banker so we can complete the transaction...

LOL....I'll call him with my Seti account password right away.
I don't want to miss such an opportunity!

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Can you order me one as well please Mark - you'l have to send your bank details to my Nigerian banker so we can complete the transaction...

LOL....I'll call him with my Seti account password right away.
I don't want to miss such an opportunity!

---

ROFLMAO, Zman!
That's the best one I've seen in a long time!

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

I'm getting old. What were we talking about again?

So am I, and for the second time this month, we're having a warm, sunny public holiday. I've just got in from a long walk with a short pit stop, and I'll be going out again to the place of beer and food shortly. The spring /early summer weather is unprecedented.

I'll leave you with this thought: Eric said 237, six days later Credit New said 676.278 The ratio of those two figures is 2.8534936708860759493670886075949, so it's conceivable Eric gave NASA the CN value with his version 2 fiddle factor divided out.

I'll leave you with this thought: Eric said 237, six days later Credit New said 676.278 The ratio of those two figures is 2.8534936708860759493670886075949, so it's conceivable Eric gave NASA the CN value with his version 2 fiddle factor divided out.

I'll leave you with another thought. I had already done all that BEFORE I wrote the message you replied to :)

I knew it was going to be 2.85 :P

Wayback and BoincStats for you too? :)
(the "6 days later" kinda fits what I did)

Edit 2: the closest I can get to the CN claim for that date, via https://web.archive.org/web/20140327064934/https://boincstats.com/en/stats/0/project/detail/, is 676.278 TeraFLOPS for 27 March 2014 - that was another question asked in the PM.

I never saw this edit. Apparently we were doing the exact same thing at the exact same time.
According to my browser history I was on boincstats @16:34 local time (+3 UTC).

Freaky.
:)

From Richard's same PM, discussed above.

The whole thread is about the complaint that CN applies normalisation which varies (downwards) over time.

From the first sentence in the OP of this thread, thread is about troubleshooting CN (and by extension the credit system in general).
I only mention this because Grant also said somewhere, "I thought this thread was about....".
Thread is about anything anyone cares to focus on about SETI credit.

But anyway, regarding normalization:
a) If it varies downwards over time that would mean the same happened under V6. Did it? (I don't think it did. And again Astropulse is not a good candidate for inspection. A 0.44 angle range or similar would be a far better choice.)

b) Has anybody ever grouped all the angle ranges (Wiggo & rob) to figure out what min/max should be first? Because you guys keep forgetting it is mathematically impossible for all ranges from both telescopes to award the same credit per second.

We can't start troubleshooting normalization without (at least) those two data points.

We can't start troubleshooting normalization without (at least) those two data points.

Why trouble shoot something that isn't necessary?
Payout Credit according to the original definition.
All of the stated goals of Credit New are met, with none of the present issues.

---

Grant
Darwin NT

I have plotted the AR for both sources. They don't lay on top of each other, and the data is only available for GBT up to a low angle. I've not got the data here, but, from memory, it is below 0.1R (with the odd "flier" above that)
For Arecibo there are three data ranges, each with their own curve: below 0.15; 0.15 to 0.7 and above 0.7. The boundaries show as discontinuities on the plot, and the each of the curves are "not bad" fits to straight lines of different slopes and y-axis intercepts.
The bulk of the data I've got is in terms of run-times, only for CPU tasks, because the credit granted vs runtime and credit granted vs AR plots had gigantic scatter so I abandoned doing them, and they weren't helping me understand the way things were being calculated once I established what was being observed was almost exactly what I expected (I started to do the code analysis after I'd gathered a few weeks worth of data).

Apart from available time there's no reason why I didn't collect data from GPUs - one day I might do a similar run on them - I'll have to limit it to just one of my crunchers - the amount of data even my slowest GPUs produce in a day makes it a very large task, and one does need a substantial range of different AR, which we don't appear to be getting just now.

---

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

We collected a huge amount of Arecibo AR data for 'estimates and deadlines revisited' - I'll still have that somewhere, together with the graphs we plotted from them. Joe's curve is still online: Imageshack has swallowed mine, but I'll still have the originals. They confirm Rob's figures, but with slightly more detail. The time axis will have changed with the additional signal processing introduced with v7, but the break points will still be the same.

Be careful when comparing AR between Arecibo and GBT. The processing required depends on the expression of AR as multiples of the telescope 'beam width': Arecibo has a larger reflecting dish, so a more tightly focused 'beam' (actually, receiving angle, but it's easier to visualise the signal paths in reverse, like a torch).

a) If it varies downwards over time that would mean the same happened under V6. Did it? (I don't think it did. And again Astropulse is not a good candidate for inspection. A 0.44 angle range or similar would be a far better choice.)

A good idea to ask that question - It did, but the slope was so low that nobody really noticed it. It would be very difficult to do the comparison between v6, v7 and v7 because each of these version introduced new techniques into both the algorithm and its implementation. I do remember that at one of the version changes there was a fair old fuss about the amount of credit awarded ("halved") and yet execution time had increased ("doubled"). As I explained some time ago the credit calculation is susceptible to sudden changes such as changing the basic calculation (v6-v7, v7-v8). This will initially be seen as an over (or under) shoot in the credit awarded, it then wobbles around a bit before settling on a new slope (which is caused by another part of the way credit is award) - and this slope can only ever have a negative gradient, and each time it suffers the big step change the slope will increase (thankfully it is still quite small, even if now noticeable)

---

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

Richard - Thanks for posting the link to Joe's curve. It looks very similar in shape to those I've plotted.
I've got more data in the 0.05-0.5 range for Arecibo which (from memory) shows the shape of that bit of the curve more clearly.
If one does the same for data from the GBT the "splatter" at the left-hand end extends to lower ARs, which is hardly surprising given the data we are analysing. Obviously the time axis is different - we used different processors and (probably) different applications.

For the purposes of this discussion I would park the AP data for now - while it appears to behave in a similar manner to one of the sections in Joe's plot, recently there hasn't been enough of it to properly analyse. It's no good collecting data from different people, one needs a single source for the pass to properly identify the "controlling factor" in the data, most people think it is "% RF pollution", but it could be something else that we haven't considered or noticed.

---

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

What I did with Joe was to identify a small number of reasonably powerful machines (possibly servers) running stock apps - I think we started looking from about 1,000 downwards in the 'top computer' rankings - that kept the optimisers out of the picture. Nowadays, you'd have to start lower down to bypass the GPU users as well. We had the advantage of an Excel script by JD Whale which enabled us to scrape the results - including AR - automatically. The result page format has changed since then, so the script wouldn't work without adaptation.

The point about the time axis wasn't that the absolute values will have changed - that's a given - but the relative runtime of the different groups will have changed, too.

"controlling factor" in the data, most people think it is "% RF pollution", but it could be something else that we haven't considered or noticed.

For AP it's%of blanking.
Saying that not from experimental data but from processing code knowledge.
Instead of MB with quite complex computational density dependance from AR (on different AR different algorithms enabled and different processing compexity inside single algorithms as well) AP is quite straightforward.
Same operations perfomed on data blocks with single exclusion - blanking areas. And cause our joint effort leaded by Joe resulted in changing from pseudo-random blanking to area skipping (in client) more blanking just means more data chunks to be skiped - simple linear dependence of computing complexity from % of blanking per task.

Regarding AP as "outlier": in no way. It's just good example HOW MUCH particular mix of FLOPs can matter and HOW MUCH memory architecture can matter.
AP operates bigger data chunks (hence issues with L1 cache on CPUs) but does mostly additions - cheapiest of FLOPs (FLOPs are not equal as was said before, and range of this inequality - ORDER of marnitude, for reference: http://ithare.com/infographics-operation-costs-in-cpu-clock-cycles/ - simple but very intuitive; also shows how much depends on memory - TWO orders of magnitude! So, FLOPs mean nothing versus good memory accesses).
So AP not fit in CPU very nicely but better suits for GPU parallelization. Hence so big difference between CPU and GPU runtimes for AP.
Taking into account that original CPU AP was wounded by algorithmic inefficiencies (for example, type cast (quite costly operation) in inner loop - how would one express it in FLOPs - no idea, I can say it's costly in CPU cycles!) and CreditScrew attempts to recalibrate on that version we get much higher credits for AP than for MB.

Could play Pythia and say that if initial AP CPU version would removed from SETI servers completely we would get huge drop in AP credits awards. No matter that our computing abilities remain exactly same. Even more, no matter that actually almost nobody uses that stock version currently cause there are much more speedy SSE ones (and again, "SSE" here only the label, they are from opt source tree and include all Joe's, my and other algorithmic optimisations as well, SIMD level only one of them) and servers send tasks for obviously better app in AP case.

That's just example how "fair" CreditScrew could compute credits.
SImple - it can't. It estimates on wrong basis.
And special note regarding how good CreditScrew as FLOP-counter or FLOPS counter.
Cause credit awarding linearly linked with FLOPS estimation by CreditScrew same applies to FLOPs and FLOPS counting abilities - CreditScrew can't count them correctly, it estimates on wrong basis.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

re APs - Thanks Raistmer - one less task on the list :-)
The Infographic is very enlightening, and is of course "aimed" at "intel-style" processors, and may not apply to ARM or other RISK processors (and almost certainly not to DSPs and GPUs.). That said it does illustrate the impact of doing things differently - which is what a lot of optimisation is about.
As a rough guide - if it's bad in clock cycles then it is almost certain to be bad in FLOPs.

Richard - I didn't explain what I meant when talking about the time axis, thanks for filling in that hole.

---

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

Ok Alex, I'll let you in on a wintertime experiment of mine.

Last winter this machine attained a RAC of 15K after 2 weeks while this year at the same time it's only got to 12K (still using the same apps and settings as last year, but likely more AP's have been processed so far this winter) and after 8 weeks it had hit 22K, but I doubt that it will get that far this year let alone reach an average of 23K before being shut down again.

Ok things have had enough time to settle down now to do a couple of comparisons since last winter here.

My main rig last winter had an average RAC of 47K, but other than the recent addition of Arecibo VLAR's to GPU's nothing has been changed, though today its RAC is down to 32K.

My backup rig Last winter had a RAC of 46K, but this winter that's down to 30K this winter.

Now my "Borg" (which has been a regular wintertime heater here) last winter averaged a RAC of 22-23K, but this winter can barely keep its head about 14K today.

Now these numbers for last winter can be verified by other members of team [url=]Australia[/url] or by mods here (or here for slightly down numbers that all can see) and the next rig mentioned.

Which brings us to this rig (which last winter was my heating backup) and averaged a RAC of 7K, but that has been replaced by this rig this year with faster quad core CPU, faster memory and dual GTX 550Ti's which hit a peak of 11K, but has now dropped well below 10K (and showing no signs of stopping yet).

I'm sorry Alex. the same apps and settings are still in use today as last winter, but CreditScrew is screwed and while some may have joked about us getting down to zero (the below zero numbers were suppose to have been fixed long ago) they may not have be far wrong in the end.

Cheers.

Thanks Wiggo - your figures support my impression that there had been a ~30% reduction in RAC over the year. Some of this can be explained by the change in the balance of Arecibo to GBT work (including the resultant dearth of AstroPulse work), but some is associated with the predicted decline in credit per equivalent task due to the way credit is granted.

---

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