HostID 1001562 - Richard Haselgrove's Q6600 Quad Core

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

RE: I've leave it to wiggle

Message 78957 in response to message 78956

Quote:
I've leave it to wiggle specialists to interpret variation near the minima in 291.65 as to whether any wiggle behavior is discernible within the noise.

At these lower signal search frequencies, the cycle period is shorter and a join-the-dots is thus more granular. We saw this in R3 initially as a consistent 'step' feature at ~ 1/4th and ~ 3/4ths of a cycle. Then later as the run progressed to higher frequencies more points exist per cycle ( longer period ) and we could discern 'beat' envelopes within long sequences of successive cycle times. This was more easily seen near runtime minima, as sinusoid curves have relatively more points near their extremes.

Quote:
I'm afraid we may not be seeing again anywhere near this many cycles of a single frequency form a single host. The pattern of issue seems to be heading toward issuing work units from sequence numbers from a small fraction of a cycle, then bumping up the frequency by .05 or more, and often repeating in the same small cycle fraction.

Drats. :-)

Quote:
I'm not ready to discuss cycle length vs. frequency, from which you may correctly infer that it does not seem to be fitting so tidily as I would like.

Ditto. :-)

Cheers, Mike.

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

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,046,141,395
RAC: 980,123

Here's a run I did on my host

Here's a run I did on my host 1574982.

343.15,81,28684.65,6.04,1574982
343.15,74,31311.12,6.04,1574982
343.15,57,27939.77,6.04,1574982
343.15,55,27973.41,6.04,1574982
343.15,54,28058.59,6.04,1574982
343.15,53,27792.37,6.04,1574982
343.15,52,27897.77,6.04,1574982
343.15,51,27924.82,6.04,1574982
343.15,50,27965.24,6.04,1574982
343.15,49,27993.16,6.04,1574982
343.15,48,28255.40,6.04,1574982
343.15,47,28312.65,6.04,1574982
343.15,46,28439.11,6.04,1574982
343.15,45,28504.78,6.04,1574982
343.15,44,28917.34,6.04,1574982
343.15,43,29473.91,6.04,1574982
343.15,42,29679.24,6.04,1574982
343.15,41,30295.78,6.04,1574982
343.15,40,30446.95,6.04,1574982
343.15,39,30965.36,6.04,1574982
343.15,38,31344.83,6.04,1574982
343.15,37,31793.03,6.04,1574982
343.15,36,32243.11,6.04,1574982
343.15,35,32114.76,6.04,1574982
343.15,34,31720.23,6.04,1574982
343.15,33,31368.59,6.04,1574982
343.15,32,30803.69,6.04,1574982
343.15,31,30568.38,6.04,1574982
343.15,30,30092.55,6.04,1574982
343.15,29,30098.08,6.04,1574982
343.15,28,29306.75,6.04,1574982
343.15,27,29253.42,6.04,1574982
343.15,26,28815.76,6.04,1574982
343.15,25,28411.62,6.04,1574982
343.15,24,28239.19,6.04,1574982
343.15,23,28166.12,6.04,1574982
343.15,22,28301.28,6.04,1574982
343.15,21,28070.38,6.04,1574982
343.15,20,27997.14,6.04,1574982
343.15,19,27940.26,6.04,1574982
343.15,18,28049.40,6.04,1574982
343.15,17,27946.14,6.04,1574982
343.15,16,28090.41,6.04,1574982
343.15,15,27981.15,6.04,1574982

This is machine is used for the ordinary day to day activities.

I'll leave the maths to you guys, I'm not to good at that.
Hope it helps.

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

Here's just some trials using

Here's just some trials using RR7F, using Peter's sky grid density constant estimate of 0.000325 ( single cycle equivalent mappings shown ):

For Richard:

and Holmis:

so clearly there's trouble at mill, for RR that is, particularly with the cycle period.

Cheers, Mike.

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

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,046,141,395
RAC: 980,123

I have also tried with "RR7"

I have also tried with "RR7" but I used a constant of .00029 on the data from my host.

Not a perfect fit, but pretty close.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,020
Credit: 4,979,732,535
RAC: 3,064,949

RE: I have also tried with

Message 78961 in response to message 78960

Quote:

I have also tried with "RR7" but I used a constant of .00029 on the data from my host.

Not a perfect fit, but pretty close.

My last try a couple of days ago I was using a three parameter model, with a linear adjustment each on the cycle count and the frequency to twist the original quadratic a little. As it happens, my last try I had:

quad parm 0.00029
cycle adder 1.2
freq adder 1

I don't like adding so many knobs--it suggests the model is structurally wrong, and I don't have enough data to fit this many parameters with any confidence.

Also, too little of my data comes from actual observation.

I certainly agree that my first guess of .000325 on the original quadratic is too high, and gets more so with higher frequencies.

Two other comments on your host's data:

My observation of increasing CPU time with cycle count is not in evidence (the minima near 18 and 54 look matched).

Your noise level is much less than mine, suggesting much of the small-scale variation on both my hosts comes from application environment variation, not the Einstein ap itself.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,020
Credit: 4,979,732,535
RAC: 3,064,949

RE: My last try a couple of

Message 78962 in response to message 78961

Quote:
My last try a couple of days ago I was using a three parameter model, with a linear adjustment each on the cycle count and the frequency to twist the original quadratic a little.


So I had another try just now at making a cycle-length estimation function for S5R4:

My latest try went back to something I tried but seemed not to help on S5R3--putting 10 Hz steps into the relationship (as suggested by the actual step of switching from one skygrid file to the next)

So my latest try at a fit with more knobs than I like is:

cycle length(freq)=0.42+.00029*(ceiling(freq,10)+1)^2

(where the values .42, .00029, and 1 all are just fitting parameters, not taken from anything remotely fundamental and the ceiling function is expressed in the MS Excel style).

Please note that even if the use of the ceiling function is actually an improvement, that the actual steps would likely be at the changeover point to the next higher skygrid, which is not precisely at a 10 Hz boundary (I actually took the trouble to observe it directly once, but I don't think I kept my notes, anybody else got that?).

Caveats aside, this actually looks pretty close to the Bikeman-style estimates (skygrid line count divided by skypoints) for nine frequencies between 180 and 970 with a maximum error of 0.49, and estimates based on actual observation for five frequencies between 180 and 402 with a maximum error of .55. The error ranges and maximum errors in both cases are better than I managed to twiddle parameters to without the 10 Hz step, which is slightly confirmatory.

I am less far along on possible issues of CPU time increase at same frequency with cycle count (observed on my hosts, but not confirmed by the data provided by Holmis) and increase with frequency. It seems unlikely that I'll get enough data to provide a reasonable approximation for either of these anytime soon from observation. If improvement in this came from algorithmic understanding, that would be wonderful.

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,046,141,395
RAC: 980,123

I had some luck and got a

I had some luck and got a sequence at the third cycle minimum and these are my data:

434.30,166,31342.62,1574982
434.30,151,28367.81,1574982
434.30,150,28112.16,1574982
434.30,149,27810.78,1574982
434.30,148,27937.05,1574982
434.30,147,28075.33,1574982
434.30,146,28007.25,1574982
434.30,145,27935.06,1574982
434.30,143,28141.98,1574982
434.30,142,28214.24,1574982
434.30,141,28218.77,1574982
434.30,140,28066.89,1574982
434.30,139,28131.44,1574982
434.30,138,28281.35,1574982
434.30,137,28241.29,1574982
434.30,136,28066.42,1574982
434.30,135,27817.40,1574982
434.30,134,28039.06,1574982
434.30,133,28372.60,1574982

When using RR7 and setting "Peter's sky grid density constant" at .00029 i see some serious wobbles!

Edit: I do have task with sequence numbers from 132 down to 94 in cache if it will help, but it will take a few days to process them...

archae86
archae86
Joined: 6 Dec 05
Posts: 3,020
Credit: 4,979,732,535
RAC: 3,064,949

RE: Edit: I do have task

Message 78964 in response to message 78963

Quote:
Edit: I do have task with sequence numbers from 132 down to 94 in cache if it will help, but it will take a few days to process them...

In that frequency I'm already pretty confident that the "real" cycle length is 57, but it would be good to review your host's data when results from about 108 through 120 are available, as a check.

I am particularly in the market for any cases in which data are available from both sides of a peak for frequencies above 500 Hz, the higher the better. It would be particularly wonderful if in such a case I could get the line count of the associated skygrid file as well. I don't need a perfect sequence of consecutive numbered results. As I'm not a "wiggle" guy, data near the minimum is far less helpful to me than data on either side of a maximum. Alternately, data from the same slope of two different maxima can be pretty helpful (I've actually seen this once, and it accounts for my only observation for a cycle time for a frequency above 402 Hz.

At lower frequencies, the "Bikeman" rule (dividing skygrid line count by skypoints) is generally giving cycle length estimates that are an integer to a resolution of .01. Further, in the half dozen cases where I've been able to compare observation with that rule, the match has been pretty close. However, I have only one observation above 402 Hz, and that is just 623 Hz, so I'm hoping to get some more. Also, up there the Bikeman rule no longer gives results particularly close to an integer.

RandyC
RandyC
Joined: 18 Jan 05
Posts: 3,711
Credit: 111,139,797
RAC: 0

If anyone is interested in

If anyone is interested in harvesting some mostly sequential data, take a look at this host. I've had a fairly long string going and it looks like I'm still being issued WUs off the same data pak.

I've even had some resends pop up in the middle.

Seti Classic Final Total: 11446 WU.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,041
Credit: 741,878,982
RAC: 1,419,963

Been a while since we had one

Been a while since we had one of these:


(direct link)

As others have noted, there seems to be a change in work generation/allocation this run. I seem to be getting similar sequence numbers from a range of frequencies, rather than a range of sequence numbers from the same frequency - apart from the spludge in the top-rught corner, I've also got a number of 1590-ish sequence numbers from the intermediate frequencies missing from this list. Altogether, there's a lot of data file deleting and downloading going on.

But I got this one nice run. Not enough to nail down the maxima and minima, but plenty of food for thought for all you curve-fitters out there.

Comment viewing options

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