HostID 1001562 - Richard Haselgrove's Q6600 Quad Core

Astro
Astro
Joined: 18 Jan 05
Posts: 257
Credit: 1,000,560
RAC: 0

OFF TOPIC post ahead Water

OFF TOPIC post ahead

Water pipe burst on a weekend. How typical. You do know why most pipes pick the weekend to burst don't you? (especially for those on a public water supply). It's because factories shut down on the weekend and water pressure rises.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,493
Credit: 65,470,826,051
RAC: 54,054,376

RE: You do know why most

Message 78918 in response to message 78917

Quote:
You do know why most pipes pick the weekend to burst don't you?

Actually they don't pick the weekend - it can be any day as long as it's 3.00am - for much the same reasons you quote :-).

Actually Murphy was asleep at the wheel when this one happened. 10am on a Saturday morning immediately after breakfast has finished and the weekly washing has been hung out to dry is a much more civilised time for this event to happen. The water pipe is standard 3/4" galvanised that's been in the ground for the last 40 years. It had probably already considerably outlasted its life expectancy given that the subsoil is a pretty reactive clay. It also decided to rupture in a general area where the flow could leak into a rubble drain and become visible very quickly so we shouldn't get an enormous water bill. Yep, Murphy was definitely asleep :-).

Cheers,
Gary.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,041
Credit: 741,822,982
RAC: 1,421,067

Sorry, I've been neglecting

Sorry, I've been neglecting my charting duties too - and I haven't even got the excuse of fixing Gary's water pipe. And no more earthquakes, though we've had plenty of gales.

RH #2

(direct link)
Doing my best to get the double-helix action going again, but the scheduler won't cooperate - keeps shifting me up a band.

Peanut:

(direct link)
The original, and still the best.

Nivekfalk:

(direct link)
A close runner-up to Peanut.

Herr Datenrat:

(direct link)
I think we can ink Datenrat's Peak firmly into the atlas - even that little col at the summit is repeated. He's using very unusual hardware, but I think Bernd needs to be aware that memory contention issues are going to come into play with the new apps. Especially when we finish off with....

Elphidieus:

(direct link)
What can I say!

I think I'm going to draw this little experiment to a close. The excitement seems to have gone out of the chase, especially since Bernd came in and gave away the ending (boo! hiss!): and most of the recent additional data has merely been confirmation of trends we've already seen. So I'm putting away the graph-paper, and returning #2 to cross-project duties.

But we know we've got the technology now ready for Bernd's (or Akos's) next set of applications: and if anyone spots an interesting host, upload the data and I'll chart it.

TTFN.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,493
Credit: 65,470,826,051
RAC: 54,054,376

RE: Sorry, I've been

Message 78920 in response to message 78919

Quote:
Sorry, I've been neglecting my charting duties too

Richard, thanks very much for the latest plots which show both the "archae86 cycles" and the "Haselgrove wiggles" to perfection. I realise that it would have taken you a lot of time and effort to produce the plots and all I can say is that I really appreciate it that you made the effort.

Quote:
I think I'm going to draw this little experiment to a close.

Yes, you have really produced enough at this stage to show how complex a problem it really is. With the magnitude of the wiggles near the troughs it's little wonder that Mike's RR (as it currently stands) is not able to handle the "wing flaps" as I think he called them. For my part, since it takes no real effort, I'll continue gathering and uploading data so that Mike has a continuing resource to use if he wishes. I'll do it less frequently and I won't bother to announce it unless something special happens.

Quote:
The excitement seems to have gone out of the chase, especially since Bernd came in and gave away the ending (boo! hiss!): and most of the recent additional data has merely been confirmation of trends we've already seen. So I'm putting away the graph-paper, and returning #2 to cross-project duties.

That's fair enough. I don't think it's a matter of loss of excitement but rather a realisation of the enormity of the task to overcome the modeling problem created by the wiggles. Also, in my case, there have been a few additional distractions such as converting all the fleet to run the appropriate power app and handling the havoc caused by repeated unscheduled power outages of late. The building I occupy is on the fringe of the CBD and there are quite a few major redevelopments occurring in the precinct. There have been at least 4 power outages caused by contractors digging up cables, etc, in the last month alone. The latest one was yesterday and it lasted about 1.5 hours. After the power comes back on, the circuit breakers usually blow, not because all machines are trying to fully restart but more because of powering up the transformers to provide the standby voltages. I usually have to pull most of the power cables to allow the circuit breakers to be reset. Then it takes me many hours to recover and get all machines back on line.

Quote:
But we know we've got the technology now ready for Bernd's (or Akos's) next set of applications: and if anyone spots an interesting host, upload the data and I'll chart it.

Yes, exactly, and thanks for the offer. There'll probably be a round of making some test apps "official" and then there'll probably be a fresh round of power apps to play with. Should be interesting :-).

Cheers,
Gary.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,493
Credit: 65,470,826,051
RAC: 54,054,376

Here is another update for

Message 78921 in response to message 78916

Here is another update for the 8 hosts shown below.

1001562 belonging to RH (#1).
1001564 belonging to RH (#2).
997488 belonging to peanut.
1084360 belonging to nivekfalk.
1101980 belonging to RAMA.
1106990 belonging to Herr Datenrat.
989926 belonging to Elphidieus.
1109009 belonging to th3.

Since Richard is now switching his #2 machine to a mix of projects and since the nature of the cycles and wiggles has been well documented, I'm not going to continue with as many updates or even with reporting them. I'll still do the updates probably about twice a week so as to maintain the data resource in case anything of interest comes along. Mike will at least have a growing archive of data if he needs it.

As Richard put it, the systems are all in place should a sudden event of interest emerge :-).

Cheers,
Gary.

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,219
Credit: 137,245,202
RAC: 54,848

Please do continue chaps, if

Please do continue chaps, if you have the inclination, as your efforts have certainly held my interest. :-)

Examining the lower part of the curve, if one subtracts away the general sine character ( this gives the 'residuals' after a least squares fit ), then it resembles a 'beat' waveform. You get this when waveforms with two ( slightly different ) frequencies are combined. This yields a variation in this combined curve with a fast rate ( the rapid wiggles ) corresponding to the sum of those frequencies, and a variation 'envelope' with a slow rate ( the wax and wane around the base sine curve ) corresponding to the difference of those frequencies. I think this image illustrates this well, and I think the 'wiggle envelope' persists to the peaks.

For instance if you combine ( with equal amplitudes say ) 99Hz and 100Hz signals then you'll get 199Hz wiggles modulated by an envelope that rises and falls at a 1 Hz rate.

The frequencies in question refer to the variation across sequence number, of some two parameters in the algorithm that significantly determine runtime. I think Bernd's comment in this post is apropos :

Quote:
So the current run-time is the sum of a part that is determined only by the number of templates in each task (and can be derived from the currently assigned credit), and a part that varies with the (average) declination of the sky positions in the sky partition assigned to the task.

So this may give us two further degrees of freedom to model against - a more elaborate least squares ought to do it, as one can use least squares in any number of dimensions ( limited by exhaustion ). Dear old Gauss used quite a few when trying out planetary orbital fits .... :-)

I too have had some real-life distractions, not plumbing or power thankfully, but exams etc ...... which went well though, however I dreamt eigenvectors for a week after :-). I'm now working on RR8 - a full Java version - and have just this evening cleared a tricky coding logjam ( aka a wrestle to the death ) related to Java's confusing syntax with respect to generic types, abstract classes and interfaces [ for those of the knowledge : Java doesn't do multiple inheritance of classes like C++, and kludges therefore abound ]. This will ensure reliable, and dare I say foolproof, cross-platform data handling of any program inputs. I foresee that the GUI coding ( which I may have a surprise or two ) will be much easier having nailed the underlying representations in a type-safe manner.

So it's more a drawing of breath, than a loss of the chase .... :-)

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,219
Credit: 137,245,202
RAC: 54,848

Yeah, the more I look at it

Yeah, the more I look at it the more convinced I get! :-)

Indeed the '0.45' thing mentioned quite a while ago ( I think Gary first mentioned that as such ), the 'stutter' half way up the climb is where the envelope compresses to zero width within a few sequence numbers. Then there's a short run of rapid swings prior to a smooth stretch to the top.

From one peak to the next, there seem to be six 'nulls' where the envelope is thinnest. This means we have 3 full cycles of the envelope for ~ 515 - 345 = 170 sequence numbers. Or a + b = 170/2 = 85, whereas a - b = 3 -> giving a = 44 and b = 41. So ( sinusoid aside ) you have two factors which vary in the ratio ~ 44:41. That is the beat of two close rhythms during a pole to pole sky traverse, and 44:41 = 88:82 = 172:164 - so that fits with the idea that the template placements don't exactly match integral degrees!

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,493
Credit: 65,470,826,051
RAC: 54,054,376

RE: Please do continue

Message 78924 in response to message 78922

Quote:
Please do continue chaps, if you have the inclination, as your efforts have certainly held my interest. :-)

I've uploaded a further update of the 8 hosts. All have quite a few extra results on top of what was there last time I updated.

Cheers,
Gary.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,041
Credit: 741,822,982
RAC: 1,421,067

Bumping this thread so we can

Bumping this thread so we can recycle it for the S5R4 timing run I've just started.

Here is the chart from identical hardware, using power app 4.36 for S5R3 last time.


(direct link)

A chart for frequency 0256.50, tasks __16 to __0, using SSE app 6.04_1 for S5R4 (that should have directly comparable timings) will appear here on Sunday afternoon/evening, UK time, if nothing goes wrong.

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,219
Credit: 137,245,202
RAC: 54,848

Thoughts/questions : #1.

Thoughts/questions :

#1. The runtimes ( all else equal, given machine config etc, b/w S3 & S4 ) should increase in general proportion to the greater integration time ( ie. true IFO collection time )?

#2. Effect of this weighted Hough business - ie. accounting for the IFO 'antennae sensitivity' w.r.t direction - which way? More efficient search? Less efficient search? Or ~ same efficiency but with other benefit(s)? Moving the weights in to E@H units eases post-processing?

#3. The wiggly bits. My whoopsie. The beats I referred to a while go like (f1 + f2)/2 and (f1 - f2)/2 in absolute terms, but the ratio b/w those is the same. The f's here are frequencies inherent in the quasi-sinusoid behaviour/pattern of runtime vs. sequence number NOT GW signal search frequency.

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter. Blaise Pascal

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.