Report on new optimized Astropulse apps for Windows

[Jason G]

I'm seeing a similar thing in your result file posted before , to my 'full blanked' result... A bunch of repeating pulses collected at the end of the WU.

My 3 results... (attached) 4.35, 5r69 and 5r69FullBlank for a WU that had lots of signals, but didn't overflow with either 4.35 or v5,

Note full blank repeating pulses x 30 all around <peak_bin> 33538050 give or take.  They also seem to be all around <scale> 4, so we probably can work out the average power there.  I think a step from the randomised to off-the edge or similar phenomemon might be occurring.  Exactly where does last dm get its overlap data from during dechirp? and is full blank randomising that too? or have we a nice step to raw data, zero or some averaged value?

Jason

[Note on FFT/High pass filter thing: A quick look it does *appear to me* to be the full complex fft result we're processing.  [But note that I always hated digging through buried functionality just to find what should be part of the function name, or a direct parameter] . If so that would make the frequency range -pi to +pi periodic,  low frequencies in the middle, a highpass look like an inverted 'U', what dechirp does with this I don't know yet, but later in coadds after inverse fft redundant half of *something* appears to be 'chucked out' though I'm dubious on the in-code comments as to the reasoning]

Whoops [ forgot attachment .. attached... ]

[attachment deleted by admin]

[Raistmer]

@Richard
Done too. Maybe it's pretty radical proposal, but there is plenty of SETI search for MB tasks still to probably waste resourses on app under big suspiction for now....

@Jason
LoL, comments... I liked that comments on first reading )))) But it's still unclear, did he check or not...

[Leaps-from-Shadows]

I have suspended the seven queued work units that Cruiser has for Main.  There are two in progress that I'm going to let finish, then it's up to the coding experts to decide whether the suspended units get aborted or crunched.

I'll be awaiting your determination.

[Jason G]

Personally I would continue crunching in case Berkeley need data to analyse the performance characteristics themselves.  Whether the results are fundamentally correct or not, in scientific terms they still have great value for verification, and further refinement if needed.  I don't think stopping crunching them based on the anomalies we've seen so far is necessarily the ideal situation, as they could contain important pointers toward further development, or even indicate some real, little understood phenomenon lurking in some WUs.

Just my 2 cents.

Jason
 

[Raistmer]

Sure, maybe....  or maybe not.
As usual there is no announcement what is the aim of current run and, recall situation with beta testing, I personally has no trust at all to ability of AP responsible person to correctly judge what processing power is needed for current aim. It can be just debug run, installed on few thousands of hosts, with easy. In such situation most of results will go into trash because just unneeded.
I will continue running AP offline to clear situation, but no "production" run. Wanna understand what is going on and why.

[Josef W. Segur]

I'm seeing a similar thing in your result file posted before , to my 'full blanked' result... A bunch of repeating pulses collected at the end of the WU.

My 3 results... (attached) 4.35, 5r69 and 5r69FullBlank for a WU that had lots of signals, but didn't overflow with either 4.35 or v5,

Note full blank repeating pulses x 30 all around <peak_bin> 33538050 give or take.  They also seem to be all around <scale> 4, so we probably can work out the average power there.

 

Repetitive pulses with <scale> 4 are from the the short FFA, and although <peak_bin> shows a high number they are actually working on data from the first 1M samples. Subtract 33538048 from the <peak_bin> value to correct it. Even then, it looks like the remaining value needs to be scaled up by 2^(scale + ffa_scale).

<scale> 4 says the power was x16 when copied to the FFA buffer, if there's any non-zero <ffa_scale> value those are also doublings. But the actual folding also increases power by a factor between 137 and 274, it should be possible to figure that out from the <period> but I haven't looked at that closely enough to know what units it's using.

I think a step from the randomised to off-the edge or similar phenomemon might be occurring.  Exactly where does last dm get its overlap data from during dechirp? and is full blank randomising that too? or have we a nice step to raw data, zero or some averaged value?

Jason
...

For "blanking", a 32K pseudo-random data sample chunk replaces 32K real samples before the forward FFT, there's no chance that any deeper parts of the single pulse processing will see a mix.

There IS a flaw in copying data to the FFA buffers, though. Because the data chunks are overlapped by half, only the first half of the coadded chunk is copied, then when the last chunk to be copied for a dm is reached there's half a chunk worth which needs special handling. Rather than putting the last half of the chunk there, the code puts a second copy of the first half. That's wrong and should be corrected, but I doubt it is contributing much to the weird results.
                                                                    Joe

[Leaps-from-Shadows]

Cruiser's compiled results so far:

Main:
368768645 (Validated) - 0.008021076 credits per CPU second
368768653 (Validated) - 0.008183099 credits per CPU second
368496971 (Pending) - 0.008235288 credits per CPU second
368768654 (Validated, canonical result) - 0.008136883 credits per CPU second
368768675 (Pending) - 0.008161555 credits per CPU second

368418713 (v4.37, Pending) - 0.006726898 credits per CPU second

Beta:
1563494 (Validated) - 0.009484371 credits per CPU second
1565756 (Validated) - 0.009705443 credits per CPU second
1564992 (Validated) - 0.009227865 credits per CPU second
1565745 (Pending) (Task details) - 0.009846161 credits per CPU second

For comparison:
Shorty Multibeam (16.84 credits):  0.007037046 credits per CPU second
Average Multibeam (44.12 credits):  0.008375063 credits per CPU second
Long Multibeam (63.86 credits):  0.010619245 credits per CPU second

I have decided to keep crunching Astropulse units on Main, but only one at a time.

[Leaps-from-Shadows]

Cruiser's compiled results so far:

Main:
368768645 (Validated) - 0.008021076 credits per CPU second
368768653 (Validated) - 0.008183099 credits per CPU second
368496971 (Pending) - 0.008235288 credits per CPU second
368768654 (Validated, canonical result) - 0.008136883 credits per CPU second
368768675 (Pending) - 0.008161555 credits per CPU second
368768681 (Validated) - 0.008024405 credits per CPU second

368418713 (v4.37, Pending) - 0.006726898 credits per CPU second

Beta:
1563494 (Validated) - 0.009484371 credits per CPU second
1565756 (Validated) - 0.009705443 credits per CPU second
1564992 (Validated) - 0.009227865 credits per CPU second
1565745 (Pending) (Task details) - 0.009846161 credits per CPU second
1565021 (Pending) (Task details) - 0.009777995 credits per CPU second

For comparison:
Shorty Multibeam (16.84 credits):  0.007037046 credits per CPU second
Average Multibeam (44.12 credits):  0.008375063 credits per CPU second
Long Multibeam (63.86 credits):  0.010619245 credits per CPU second

[Leaps-from-Shadows]

Cruiser's compiled results so far:

Main:
368768645 (Validated) - 0.008021076 credits per CPU second
368768653 (Validated) - 0.008183099 credits per CPU second
368496971 (Pending) - 0.008235288 credits per CPU second
368768654 (Validated, canonical result) - 0.008136883 credits per CPU second
368768675 (Pending) - 0.008161555 credits per CPU second
368768681 (Validated) - 0.008024405 credits per CPU second
369656439 (Pending) - 0.007830532 credits per CPU second
349739781 (Validated) - 0.008531656 credits per CPU second

368418713 (v4.37, Pending) - 0.006726898 credits per CPU second

Beta:
1563494 (Validated) - 0.009484371 credits per CPU second
1565756 (Validated) - 0.009705443 credits per CPU second
1564992 (Validated) - 0.009227865 credits per CPU second
1565745 (Pending) (Task details) - 0.009846161 credits per CPU second
1565021 (Pending) (Task details) - 0.009777995 credits per CPU second
1565898 (Pending) (Task details) - 0.009551190 credits per CPU second

For comparison:
Shorty Multibeam (16.84 credits):  0.007037046 credits per CPU second
Average Multibeam (44.12 credits):  0.008375063 credits per CPU second
Long Multibeam (63.86 credits):  0.010619245 credits per CPU second

For the first time, I had two v5.00 work units on Main (369656439 and 349739781) that did not end with pulse overflows, and one on Beta (Task details) that did end with a pulse overflow.

[Raistmer]

this one near overflow too
single pulses: 8
repetitive pulses: 27

Pity that we haven's % of blanking in stderr too....

[Richard Haselgrove]

..
Oh, and please contribute to to my 'dilemma' thread at SETI
..

Done.  Feel free to disagree.

And now the WU is done, too. I decided to let it run with opti v5.00: it validated the v5.00 stock, and the v4.36 stock got zilch. Not even weakly similar. Similar situation for the first of Geek@Play's dilemma cases, which has now reported. I can't say I'm surprised, but I am disappointed. I very much doubt Eric is likely to waste his holiday weekend writing a manual credit script - and the results will be purged by then, anyway.

[Raistmer]

Well, I'm more concerned by the meaning (or meaningless) of current AP work than credits actually...

[Leaps-from-Shadows]

Cruiser's compiled results so far:

Main:
368768645 (Validated) - 0.008021076 credits per CPU second
368768653 (Validated) - 0.008183099 credits per CPU second
368496971 (Pending) - 0.008235288 credits per CPU second
368768654 (Validated, canonical result) - 0.008136883 credits per CPU second
368768675 (Pending) - 0.008161555 credits per CPU second
368768681 (Validated) - 0.008024405 credits per CPU second
369656439 (Pending) - 0.007830532 credits per CPU second
349739781 (Validated) - 0.008531656 credits per CPU second
369656217 (Pending) - 0.008278782 credits per CPU second

368418713 (v4.37, Pending) - 0.006726898 credits per CPU second

Beta:
1563494 (Validated) - 0.009484371 credits per CPU second
1565756 (Validated) - 0.009705443 credits per CPU second
1564992 (Validated) - 0.009227865 credits per CPU second
1565745 (Pending) (Task details) - 0.009846161 credits per CPU second
1565021 (Pending) (Task details) - 0.009777995 credits per CPU second
1565898 (Pending) (Task details) - 0.009551190 credits per CPU second
1628888 (Pending) (Task details) - 0.009486180 credits per CPU second

For comparison:
Shorty Multibeam (16.84 credits):  0.007037046 credits per CPU second
Average Multibeam (44.12 credits):  0.008375063 credits per CPU second
Long Multibeam (63.86 credits):  0.010619245 credits per CPU second

[Leaps-from-Shadows]

Cruiser's compiled results so far:

Main:
368768645 (Validated) - 0.008021076 credits per CPU second
368768653 (Validated) - 0.008183099 credits per CPU second
368496971 (Pending) - 0.008235288 credits per CPU second
368768654 (Validated) - 0.008136883 credits per CPU second
368768675 (Validated) - 0.008161555 credits per CPU second
368768681 (Validated) - 0.008024405 credits per CPU second
369656439 (Validated) - 0.007830532 credits per CPU second
349739781 (Validated) - 0.008531656 credits per CPU second
369656217 (Pending) - 0.008278782 credits per CPU second
370050505 (Validated) - 0.008334832 credits per CPU second
370050504 (Validated) - 0.008610554 credits per CPU second

368418713 (v4.37, Pending) - 0.006726898 credits per CPU second

Beta:
1563494 (Validated) - 0.009484371 credits per CPU second
1565756 (Validated) - 0.009705443 credits per CPU second
1564992 (Validated) - 0.009227865 credits per CPU second
1565745 (Validated) - 0.009846161 credits per CPU second
1565021 (Pending) (Task details) - 0.009777995 credits per CPU second
1565898 (Pending) (Task details) - 0.009551190 credits per CPU second
1628888 (Validated) - 0.009486180 credits per CPU second
1565146 (Validated) - 0.009490379 credits per CPU second

For comparison:
Shorty Multibeam (16.84 credits):  0.007037046 credits per CPU second
Average Multibeam (44.12 credits):  0.008375063 credits per CPU second
Long Multibeam (63.86 credits):  0.010619245 credits per CPU second

[[B^S] zioriga]

Leaps-from-Shadows excuse me, but if you use Vista x64 why don't you use BOINC 6.2.19 64b client ??