S5R3 beyond 800 work
30 Jan 2008 20:12:58 UTC
Topic 193480
(moderation:
I thought it might be good to have a thread dedicated to discussing the Einstein S5R3 work at frequencies above 800.
I've only seen 800+ work on one host. It picked up frequency a 818.50 result 30 Jan 2008 0:49:38 UTC. I assumed because of the timing that it was one of the new build, but since the host Workunit was distributed to my quorum partner on January 16, and I validated against his, I think instead this is an indication that catchup results for the bad "easy-runners" are still getting distributed in cases where one of the original quorum partners failed to return on time or ones for which a long succession of client and download errors has run the clock until now. The run time on my host was indeed considerably less than prior experience and the cyclic dependence would have made me expect.
So for those posting 800+ timings to this thread, please do some checks to distinguish between the "good, long, new" 800+ and the "bad, short, old" ones. Initially I suggest checking the work unit for oldest issue date--possibly there will be some more definitive marker, perhaps the string "S5R3b" will appear somewhere.
On my little fleet of four machines, I don't think I've seen a post-fix 800+ result download yet.
S5R3 beyond 800 work
)
Very good idea. Also, I agree that the ones you have so far are all from the old series. Hopefully the cleanup will be fairly quick and we wont see too many more of these. I've got several machines that had +800 work - I've got 810, 820 and 830 skygrids - but all of this was sent before the problems were announced. One of the machines I'm saving data on has a good string of sequence numbers before switching back to a 7xx frequency when the +800 stuff was cancelled. I was just going to ignore these "fast" results but I can provide the information if anyone is crazy enough to be interested :).
If you check the server status page you will see that the WU generator has not been restarted yet and we still have ~460K unsent results. I imagine they may let this number decay for a while before producing any +800 work. Unfortunately this makes it difficult to collect good strings of single frequency sequence numbers as the frequency keeps changing for new tasks during the dregs cleanup.
Cheers,
Gary.
Ah, Bernd has recently posted
)
Ah, Bernd has recently posted that the first "good, long, new" results are being sent out, and that "The task names end in 'S5R3b', and the data files in 'S5R3'".
RE: ... the first "good,
)
That wasn't a "choose any two" scenario, was it, by any chance? :).
If so, I'll just take the "good, new" ones thanks :).
I've been checking a lot of the recent comms from my machines and I've seen changes in frequency but nothing above 800 yet.
Cheers,
Gary.
RE: RE: ... the first
)
Howdy,
hopefully there are those "good, new ones" .. I tried to get WU to crunch and the results are the following:
1/31/2008 10:51:58 PM|Einstein@Home|Sending scheduler request: Requested by user. Requesting 21852 seconds of work, reporting 0 completed tasks
1/31/2008 10:54:03 PM|Einstein@Home|Scheduler request failed: HTTP internal server error
1/31/2008 10:55:03 PM|Einstein@Home|Sending scheduler request: Requested by user. Requesting 21868 seconds of work, reporting 0 completed tasks
1/31/2008 10:56:38 PM|Einstein@Home|Scheduler request failed: HTTP internal server error
( Well, I got a new WU just now when typing this ( h1_0711.20_S5R2_55_S5R3a_4 ) , seems to be normal ( <800 ) WU *BUT* where it's coming ?!?!? ( as the generator is *not* running now ))
-- Lundi --
RE: ( Well, I got a new WU
)
Cheers,
Gary.
Have just got five WU's in
)
Have just got five WU's in the 921 frequency range, so I will see how they go.
Here is some data for a new
)
Here is some data for a new +800 frequency task:-
If you take a look at this results list, you will see a total of 7 results at the moment. The very top two have a frequency of 891.50. One has been returned successfully and one is still crunching. The app version used was 4.26.
The next two down the list had a frequency of approx 740 and were also crunched with the 4.26 app. The three below this were crunched (wholly or partly) by the 4.15 app.
Using RR_V4 and the two 740/4.26 results, the following figures are obtained:-
The 892.50/4.26 result that has crunched has a sequence# of 493 which represents a phase of 0.01118 in its own cycle. In other words it is quite close to a peak. For the 740 frequency cycle, the closest sequence# that has a similar phase is 227 which has a phase of 0.01122. The prediction for a 740 - #227 result is 110544. This compares quite favourably with the actual crunch time of the 891.50 result of 109489 - an error of only 1%.
I think this is fairly good evidence that frequency itself, at these higher levels, has little effect on crunch times. Over the coming days as a string of 891.50 results builds up, it will be interesting to see how closely the new cycle parameters match the old.
Cheers,
Gary.
RE: Have just got five WU's
)
First result returned on an AMD Opteron 275 @2.2 GHz (standard) was this WU with a time of 50,176.77 seconds on a frequency of 0921.00 and sequence of 526.
Last two Wu's from 0747.15 frequency run were Sequence 1 with 49,607.19 seconds, and Sequence 0 with 50,438.84 seconds.
So it seems I am near a peak to start with? Very similar to lower frequencies.
All runs are on Linux with application version 4.27.
RE: So it seems I am near a
)
Yes you are very near the peak.
For the 921.0 frequency, the period of the cycle is 174.7 so that sequence numbers for peaks and troughs will be approximately:-
Peaks: 0 175 349 524 690 ...
Troughs: 87 262 437 612 ...
So your sequence number of 526 is pretty close to the 524 peak.
Cheers,
Gary.
Some run time numbers for
)
Some run time numbers for 800+ frequencies compared to sub 800 frequencies on a AMD Opteron 275 @2.2 GHz with both App 4.14 and App 4.27
Frequency...... Seq No... Run Time... Application.. Peak
0641.15............253...........51,185.24.......4.14.........254.04
0641.15............252...........49,934.31.......4.14.........254.04
0641.15............251...........49,113.63.......4.14.........254.04
0641.15............248...........47,703.01.......4.14.........254.04
0641.15............178...........49,210.59.......4.14.........169.36
0641.15............165...........49,191.75.......4.14.........169.36
0641.15............164...........48,691.80.......4.14.........169.36
0747.05............232...........51,970.29.......4.14.........229.94
0747.05............105...........48,285.97.......4.14.........114.97
0747.05.............99...........46,558.87.......4.14.........114.97
0747.15..............1...........49,607.19.......4.27...........0.00
0747.15..............0...........50,438.84.......4.27...........0.00
0921.15............526...........49,689.24.......4.27.........524.40
0921.10............526...........49,718.80.......4.27.........524.34
0921.10............525...........Yet 2 run.......4.27.........524.34
0921.10............524...........Yet 2 run.......4.27.........524.34
0921.05............526...........49,840.79.......4.27.........524.28
0921.00............526...........50,176.77.......4.27.........524.22
0921.00............525...........49,802.05.......4.27.........524.22
0921.00............524...........48,993.20.......4.27.........524.22
0921.00............523...........49,031.23.......4.27.........524.22
0921.00............522...........48,305.16.......4.27.........524.22
0921.00............521...........48,086.17.......4.27.........524.22
0921.00............520...........47,194.15.......4.27.........524.22
0921.00............519...........47,200.41.......4.27.........524.22
0921.00............518...........yet 2 run.......4.27.........524.22
0921.00............517...........yet 2 run.......4.27.........524.22
0921.00............516...........yet 2 run.......4.27.........524.22
0921.00............515...........yet 2 run.......4.27.........524.22
So run times are very similar to those in App 4.14 and 4.27 on the peaks, frequency does not make a lot of difference and neither has the App version.
I have covered the peak very well in the 0921.00 frequency run with consecutive work units.
It will be interesting to see if there are any changes in the trough sequences, in this case that wont be till sequence number 349.48 as I am heading down in the sequence run.