TAS members' thread February-March 2007: DPAD and ex-NanoHive@Home.

[caferace]

I think rabrittain must know Rudy Toody, or else it's a big coinkydink.



-jim

 

[petrusbroder]

Huh? :Q

 

[caferace]

Originally posted by: petrusbroder
Huh? :Q

rabrittain and Rudy Toody both are running FX-60's, and if you look at the rig specs on Rudy's website, the background image is by an "rabrittain".

I have an odd way of dot-connecting.

-jim

 

[petrusbroder]

Just great - but I have not checked R-T's website --- so I may be forgiven ...
Otherwise: nice observation ... now lets guess what the R.A. stands for ... :Q

 

[BlackMountainCow]

3,700 creds pending on my side ... :evil:

 

[Rudy Toody]

Originally posted by: caferace
I think rabrittain must know Rudy Toody, or else it's a big coinkydink.



-jim

You are correctomundo, Klem Kadiddlehopper!

We built 4 of these beauties which we call Blue Herons after the photo that we used as the wallpaper. Isn't it terrific?

Rex doesn't know it yet, but I ordered parts for blueheron5 and he will be involved.

--Fred

 

[caferace]

New News:

Looks like (from my back-of-the-napkin calculations) that the devs just knocked another ~20% off average claimed. Getting closer to reality.

edit: They're likely on a 12 hr. cycle, given the last time I noticed the change.

-jim

 

[BlackMountainCow]

Hey Rudy, do you know Sarah Orne Jewett?

 

[Rudy Toody]

Originally posted by: BlackMountainCow
Hey Rudy, do you know Sarah Orne Jewett?

I'm not that old. She was a writer who died in the early nineteen hundreds. Why do you ask?

 

[BlackMountainCow]

Well, as you obviously like herons, I though you might like this one, too.

 

[BlackMountainCow]

[spam]

:evil: And there goes post number 4,444 :evil:

[/spam]

 

[caferace]

Originally posted by: caferace
New News:

Looks like (from my back-of-the-napkin calculations) that the devs just knocked another ~20% off average claimed. Getting closer to reality.

edit: They're likely on a 12 hr. cycle, given the last time I noticed the change.

-jim

Definitely on a 12 hr. / - 20% schedule. ~2 hrs ago I saw another drop.

Making progress and Brian appears commited, though I couldn't pin him to a "number".

-jim

edit: It does not appear to be retroactive, which is OK I suppose. All pending and granted seem to be the same as before, just newly crunched work is adjusted.

 

[caferace]

Peter, regarding a Mac client I hear you, but you might find this host (and its owner) of interest:

http://www.nanohive-1.org/atHome/show_host_detail.php?hostid=4495

Granted, it hasn't crunched anything yet, but..

-jim

 

[petrusbroder]

Yeah, I have seen that - it is the dev of Nano-Hive@Home ... I suppose he may be working on a mac-application but has obviously not crunched anything yet ...

 

[Rudy Toody]

Hi guys!

What's the word on the 17 Feb decision? Is it worthwhile staying? I've got one PC that I should set to get more work in about a day.

I'm taking one box off line in a few hours to replace the psu and to add a water block to the gpu. It should be back on line in a couple of days.

--Fred

 

[petrusbroder]

I am still evaluating the credits. But something has definitively happened:

Date ________ number of WUs ______ credit/hour ______ credit/hour ___ credit/hour
_______________ in stats ___________ average ________ std-dev ______ min - max
February 6 _______ 19 ________________ 48 _____________ 15 ________ 31 - 71
February 7 _______ 18 ________________ 46 _____________ 23 ________ 21 - 92
February 8 _______ 17 ________________ 38 _____________ 10 ________ 23 - 52
February 9 _______ 17 ________________ 34 ______________ 7 ________ 21 - 40

I do not have enough WUs in for analysing the data for February 10
All calculations are based on "granted credit" - pending credits are not included!
The highest number 92 credits/hour is not an outlier because it is still within the limits of average + (2* std-deviation)

The interesting part is that the variation between the WUs' credit/hour is decreasing so much: from 15 - 23 to 7 - 10. This indicatees that the system is getting more fair - i.e. we get the same credit for the same work performed. This is also visible in the min (credit/hour) - max (credit/hour) granted. The trend looks good!

The average credit/hour has dropped by some 30%. Not sufficiently yet, but on its way. The max-credit/hour has dropped a lot more - by more than 50%, while the min-credit/hour is more or less the same. This indicates that the adjustment process is not yet done. There are a lot of "old" WUs around and therefore the numbers will take some time to stabilize.

We see the efforts of the devs, and think we should give them next week too. I will prepare the poll during the week end and we can decide Feb. 16 - Feb-17.

 

[caferace]

Thanks for the analysis, Peter. I've had one box that has been getting a ton of short WU's, so I've been lucky enough to be able to do direct comparisons. We had another 20% drop right on schedule, 1.5 hrs ago.

I would have told you guys sooner, but I just got back from seeing Blue Man Group live. That was my third time seeing them (first in 1995) and damn, do those guys rock.

-jim

 

[petrusbroder]

Yeah, sounds good!
I'll do an ananlysis of February 10 tomorrow.
I have also seen that the short crunching results drop more (in credits/hour) than the long crunching.
It seems to me that one WU can crunch a long time on one computer and a short time on an other. on one computer the "seeds" result in a faster convergance than in an other computer, where the starting values ("seeds") start a much longer computation and result in a different minimum - similar to Rosetta@home...

 

[Rudy Toody]

Both Blue Herons are back up slinging FLOPS!

 

[petrusbroder]

Originally posted by: Rudy Toody
Both Blue Herons are back up slinging FLOPS!

Was that not a spelling mistake?

Blue Heroes are back up and slinging FLOPS!

 

[Ken g6]

Re: Macs, I assume NH has a Linux client? If so, and you're using Intel Macs, this post might interest you.

 

[petrusbroder]

Thanks.
But then I have to run Linux only - and i use my Macs for other stuff too - to much trouble to install all the different version and to re-learn and to foget the well-trained movements, menues, shortcuts etc.

 

[caferace]

I can't seem to shed my pending creds. I suppose it isn't a bad thing, but I currently have 471 WU's pending in the queue. And the pending total is *more* than it was the other day, 80k plus. Of course, I'm running a few more cores than I normally do, but....

-jim

 

[petrusbroder]

Here are the current stats for TAS crunching NanoHive@home.

It certainly looks very good.

This is an update on the analysisi of the ongoing change in credit-policy:

Date ________ number of WUs ______ credit/hour ______ credit/hour ___ credit/hour ______ decrease __
_______________ in stats ___________ average ________ std-dev ______ min - max _____ in credit/hour
February 06 ______ 19 ________________ 48 _____________ 15 ________ 31 - 71 _______ starting point
February 07 ______ 18 ________________ 46 _____________ 23 ________ 21 - 92 ____________ -5%
February 08 ______ 17 ________________ 38 _____________ 10 ________ 23 - 52 ___________ -20%
February 09 ______ 17 ________________ 34 ______________ 7 ________ 21 - 40 ___________ -30%
February 10 ______ 18 ________________ 23 ______________ 9 ________ 13 - 34 ___________ -52%

These are real values - i.e. granted credits for WUs. The important observation is not only the more realistic numbers for credits/hour but also the significantly decreased standard deviation, which indicates that that the credits are now more fair: the same amount of work gives approximately the same credits. Also the range of credits/hour has decreased ... before it was more than 4-fold, now it is less than 3-fold and decreasing.

Conclusion: the devs are listening to the critique and have adapted a more reasonable crediting policy. The change are - according to the forums - still going on.

 

[petrusbroder]

Here is the continuation of the analysis of the dropping credits/day in NanoHive@home!