HostID 1001562 - Richard Haselgrove's Q6600 Quad Core

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,516
Credit: 468,365,954
RAC: 58,055

RE: Observations: The

Message 78857 in response to message 78856

Quote:

Observations:
The wiggles are real.

I've got a theory on those but I need some more time to put this into text and illustrate it with visual evidence. Stay tuned.

Quote:

4.33 is a speed demon.
The cyclic variation is reduced in 4.33 (as Bernd expected), but I think it's still there - just.

Which is good by itself but it will make it harder to find a credits formula which fits all platforms and future app optimizations.

Quote:

There's a disjointedness between 909.75 and 909.95

Which is VERY interesting. It suggests that a switch to a different 10Hz skygrid file happened just between 909.75 and 909.95, doesn't it??

CU
Bikeman

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,491
Credit: 63,225,854,491
RAC: 54,308,608

RE: I was going to save up

Message 78858 in response to message 78856

Quote:
I was going to save up Peanut's chart until my 909.25 run was complete, but it's so pretty I can't resist.

Amazing stuff!! Thanks very much for publishing the graphic.

Quote:
Observations:
The wiggles are real.
4.33 is a speed demon.
The cyclic variation is reduced in 4.33 (as Bernd expected), but I think it's still there - just.
There's a disjointedness between 909.75 and 909.95

I can't get over how "non-disjointed" all the other frequencies are. It seems to indicate that there's a major change between 909.75 and 909.95 that's not there for any other frequency value. I'm wondering if this is where the step change in frequency for calculating the cycle period occurs?

EDIT:
Looks like Bikeman and I are thinking along the same lines :).

Cheers,
Gary.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,040
Credit: 686,901,812
RAC: 927,863

RE: I can't get over how

Message 78859 in response to message 78858

Quote:
I can't get over how "non-disjointed" all the other frequencies are. It seems to indicate that there's a major change between 909.75 and 909.95 that's not there for any other frequency value. I'm wondering if this is where the step change in frequency for calculating the cycle period occurs?


I thought you'd like it!

There are just a couple of data points from 909.90 in the file, and they seem to be joined up to 909.95, rather than 909.75. The single datum for 909.80 is too far up-sequence to be able to make a judgement.

But at least now we know another thing to look for. Gary, when we needed lots of data fast for a SETI calibration run, a good trick was to look through the 'top hosts' stats listing on the site and pick out ones that matched the characteristics we wanted to examine. If you could find a host that way (what are we looking for now? Cycle peaks, cycle troughs, frequencies around xx9.8x?) and upload the data, I'm happy to graph it - I just can't run your automatic data capture, because I don't have a Linux box here.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,491
Credit: 63,225,854,491
RAC: 54,308,608

RE: There's a

Message 78860 in response to message 78856

Quote:
There's a disjointedness between 909.75 and 909.95

Here's a thought on how to account for the disjointedness.

let's assume that there is a 10Hz step in the frequency for calculating the cycle period. Let's also assume that the 909.75 and 909.70 data belongs to the 910 skygrid whilst the 909.95 data belongs to the 920 skygrid. If you work out the difference in cycle period you get about 173.0 - 169.25 = 3.75.

Richards plotted points have seq#s around 450 - 460 which is therefore ~2.7 cycle periods from the start. The difference at this point (2.7 cycles) will therefore be 2.7*3.75 = 10.1 seq#s

This suggests to me that a seq# of 460 for the 909.95 data would therefore have the same phase as a seq# of 450 for the 909.75 data. Do we get much less disjointedness if we shift the 909.95 group about 10 seq#s to the left?

What do you think??

Cheers,
Gary.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,040
Credit: 686,901,812
RAC: 927,863

RE: What do you

Message 78861 in response to message 78860

Quote:
What do you think??


Ask and ye shall receive.


(direct link)

Pretty good, but not perfect. We need Bikeman's theory on the wiggles.

This graph also shows the 909.90 data.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,491
Credit: 63,225,854,491
RAC: 54,308,608

RE: If you could find a

Message 78862 in response to message 78859

Quote:
If you could find a host that way (what are we looking for now? Cycle peaks, cycle troughs, frequencies around xx9.8x?) and upload the data, I'm happy to graph it - I just can't run your automatic data capture, because I don't have a Linux box here.

OK, sounds good. My script takes about 15 seconds to capture your data and less than a minute to capture peanut's. It's hardly an onerous exercise to add a couple of extra hosts. I just went looking for Akos' host but it's not in the top 10 anymore. On second thoughts, a fast host where the owner is experimenting is probably a bad idea :).

If anyone has any suggestions for a host that is both fast and stable, I'll collect the data and Richard can produce his pretty plots.

Cheers,
Gary.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,008
Credit: 4,847,753,170
RAC: 3,346,293

RE: I can't get over how

Message 78863 in response to message 78858

Quote:

I can't get over how "non-disjointed" all the other frequencies are. It seems to indicate that there's a major change between 909.75 and 909.95 that's not there for any other frequency value. I'm wondering if this is where the step change in frequency for calculating the cycle period occurs?


Following up on the tip to look at the 10000-line result file, which I've originally been pursuing in hopes of learning a bit more about period, I noticed in the first one I examined that the first column looks rather like a frequency, and is a bit higher than the file name would suggest. I've now looked at a second, which is nominally sequence 479 for frequency 904.80.

The distribution of column values for that one resembles the first. The range of column 1 values which I assume to be frequency is:
from 904.9899428
to__ 905.0060750

Again the density of values in frequency space seems bimodal, with a much higher density from 904.98994248 to about 904.9918, then a little under a third that density on up.

While I imagine these observations may fit into the steps and wiggles puzzles somehow, I won't even guess as to how.

On my original intent to examine the third (declination?) column, for this sequence number 479 estimated to be at phase .8304 the dec values are:

0.7351975 for 3898 rows
0.7430678 for 4777 rows
0.7509381 for 1225 rows

Possibly the row average of the declination value in this result file might be a useful estimate of the "true" phase of a result? If so, might there be nothing restricting us to integer cycle lengths?

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,040
Credit: 686,901,812
RAC: 927,863

RE: If anyone has any

Message 78864 in response to message 78862

Quote:
If anyone has any suggestions for a host that is both fast and stable, I'll collect the data and Richard can produce his pretty plots.


Smoked Trout's host 1100395 (Windows 4.26) is just working its way up to 869.80 - might be interesting. I may pop some more in here if I find 'em during the 1 hour edit window.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,516
Credit: 468,365,954
RAC: 58,055

RE: We need Bikeman's

Message 78865 in response to message 78861

Quote:
We need Bikeman's theory on the wiggles.


:-)
OK, here's my theory on the wiggles, cycle periods , and everything :-)

1) Connection between runtime and skypositions:

The following diagrams are generated from actual results:

h1_0705.70_S5R2__204_S5R3a_0_0
h1_0705.70_S5R2__205_S5R3a_0_0
h1_0705.70_S5R2__206_S5R3a_1_0

Each diagram shows 10000 top candidates, within a grid of ca 1200 points in the sky. For each sky position, several parameters are tried so there are more candidates than sky-coordinates.

For this frequency, the predicted period by any of our models is close to 103, so we should be looking at the transition from the first to the second cycle:

Seq. no 204: almost at "north pole"

Seq. no 205: end of cycle, exactly at north pole

Seq no 206: beginning a next cycle, exactly at south pole.

This is strongly suggesting that the period is always an integer. The transition from one cycle to the next cycle always seems to happen at a workunit boundary.

To be continued...

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,040
Credit: 686,901,812
RAC: 927,863

I've had a look through the

Message 78866 in response to message 78862

I've had a look through the top 40, but didn't find any more working on obviously interesting data. 1079377 (Anonymous) and 1054609 (Armin Burkhardt) both have data covering the 0800.00 transition, if anyone's interested - the latter has a clear peak at 799.95__269.

Quote:
I just went looking for Akos' host but it's not in the top 10 anymore. On second thoughts, a fast host where the owner is experimenting is probably a bad idea :).


Akos is at #37 at the moment. He was down below 8000 seconds, then had a run of compute errors, and now he's back to 37000! Probably had to revert to stock to get his quota back up.....

I hope it was only the WUs he trashed, didn't blow a fuse on the box or something.

Comment viewing options

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