I remember..


log in

Advanced search

Message boards : SETI@home Science : I remember..

Author Message
Profile Dr Grey
Send message
Joined: 27 May 99
Posts: 49
Credit: 30,072,473
RAC: 26,280
United Kingdom
Message 1062749 - Posted: 2 Jan 2011, 15:33:17 UTC

I remember the thrill of when I first got involved in the SETI@home search. Back then I worked in a lab with three computers with just one of them connected to the internet. Each day, the first thing I would do was to pick up a 3.5" floppy and collect the work unit that had completed over night from each the unconnected machines and copy it across to the connected machine. I'd upload it and then download a fresh unit ready to run for the next 24 hours, so using the combined power of three top end computers I was crunching through a massive 3 units a day for 5 days a week. But of course things have moved on just a bit since then.
Since then, the amount of science done per wu has increased dramatically but the effect of Moore's Law has also been dramatic. My current machine which is now crunching with its graphics card as well as its multi-core cpu looks to be whizzing through wus at a rate of around 260 per day - a massive increase in throughput over the last decade.
Then I look at the server status page and see the incredible power of the combined SETI farm processing work units at the astounding rate of 47000 per hour which is an incredible search rate.
But hang on a minute.. according to the How SETI@home Works page Arecibo collects around 35Gb of data per day.
That data splits into 0.25Mb work units equating to around 140,000 units per day. With each unit having to be sent out at least twice that brings us to 280,000 per day. I don't know what the failure rate is, but lets say that it is as high as 50% then we are up to 420,000 units for each day's raw data collection.
That is 17,500 work units per hour to keep up with the real time data collection rate. Am I missing something here? If not, then we must be working through a backlog of raw data at a very respectable rate and at some point we will clearly reach a limit of 17,500 units per hour leaving much of our processing capacity unused.
Moore's Law is not going to stop any time soon, perhaps it is time to think again about the science that can be done or perhaps unload some of the effort being done on the server side out into the community?
Are my maths wrong or is the project massively over served by volunteer processing and if it is, what can be done with all this left over computing power?
____________

Dave
Avatar
Send message
Joined: 29 Mar 02
Posts: 774
Credit: 23,193,139
RAC: 0
United Kingdom
Message 1063090 - Posted: 3 Jan 2011, 18:06:36 UTC
Last modified: 3 Jan 2011, 18:06:59 UTC

In other words the rate at which work can be created will be the weak link in the chain...

When CPUs/GPUs become idle they will therefore move over to other DC projects :).

Profile Frizz
Volunteer tester
Avatar
Send message
Joined: 17 May 99
Posts: 271
Credit: 5,852,934
RAC: 0
New Zealand
Message 1063170 - Posted: 3 Jan 2011, 22:20:54 UTC - in response to Message 1062749.
Last modified: 3 Jan 2011, 22:21:49 UTC

I remember the thrill of when I first got involved in the SETI@home search. Back then I worked in a lab with three computers with just one of them connected to the internet.


Now I really had to laugh out loud :D Because I too remember those days. When I collected finished work units from a bunch of Linux hosts and copied them to the one host that was connected to the Internet.

As for your question: I see our compute capacity is growing faster than our communications capacity - the low compute/communications ratio is a looming issue that will inevitably cause problems.

I assume the S@H team is aware of this problem. At least I hope so ...

One idea would be to process the data more thoroughly (e.g. increased FFT sizes, better/additional search methods, etc.). Back in the days one Astropulse task ran for a couple of weeks - today, on state of the art GPUs, it finishes in 30 minutes. So we need to increase computation time ;)

Jason on the other hand had the brilliant idea to split data on the client side: Since the raw data for Astropulse and Multibeam is the same, why not transmit the 8MB only once, then use it 1) for Astropulse and 2) for Multibeam (split into n pices by the client computer).
____________
Petition against 1366x768 glare displays: http://www.facebook.com/home.php?sk=group_153240404724993

Profile Eric Rojas Chiong
Volunteer tester
Avatar
Send message
Joined: 2 May 10
Posts: 55
Credit: 338,397
RAC: 0
United States
Message 1065350 - Posted: 10 Jan 2011, 21:19:59 UTC - in response to Message 1063170.

Amazing isn't it. Personally, I hope that other software processing issue besides the Fast Fourier would be refined, such as the spectrum analysizers like the US Navy have in isolating discreet frequencies.

Larry Monske
Send message
Joined: 17 Sep 05
Posts: 262
Credit: 526,044
RAC: 42
United States
Message 1071252 - Posted: 27 Jan 2011, 21:59:51 UTC - in response to Message 1065350.

Dr Gray in my college days I also had a great fasination for SETI and still do. I helped out at the college computer lab had 37 computers running 24-7. I dont remember how much we processed but I lost all that information from the college. Im thinking we just arent in the ball park with detecting signals. I would expect some type of burst communication because of the distances involved. Most information at the fastest transmission. I wonder if in fact sometime in microwave bandwidth would work over great distance. Either that or ultralow frequencies. We are examining the middle.

Profile Dr Grey
Send message
Joined: 27 May 99
Posts: 49
Credit: 30,072,473
RAC: 26,280
United Kingdom
Message 1072862 - Posted: 31 Jan 2011, 23:07:54 UTC - in response to Message 1063170.

[quote]

Jason on the other hand had the brilliant idea to split data on the client side: Since the raw data for Astropulse and Multibeam is the same, why not transmit the 8MB only once, then use it 1) for Astropulse and 2) for Multibeam (split into n pices by the client computer).


Sounds like good thinking to me. As the tech evolves we need good ideas to make the best of the resources that become available and to overcome the bottlenecks that appear.
____________

Profile Dr Grey
Send message
Joined: 27 May 99
Posts: 49
Credit: 30,072,473
RAC: 26,280
United Kingdom
Message 1072865 - Posted: 31 Jan 2011, 23:19:28 UTC - in response to Message 1071252.

Dr Gray in my college days I also had a great fasination for SETI and still do. I helped out at the college computer lab had 37 computers running 24-7. I dont remember how much we processed but I lost all that information from the college. Im thinking we just arent in the ball park with detecting signals. I would expect some type of burst communication because of the distances involved. Most information at the fastest transmission. I wonder if in fact sometime in microwave bandwidth would work over great distance. Either that or ultralow frequencies. We are examining the middle.


I guess we have to start looking somewhere, but I'm beginning to wonder how many times we've looked at each spot in the sky for our set of frequencies now. Maybe we should be looking elsewhere too? I would also be interested in seeing the figures for the ratio for the rate of data collection to the rate of data analysis. I estimated that we could be close to 1:1 now and the receiver seems to be off for much of the time as well.
Is now the time to start looking at other frequencies as well?
____________

Profile soft^spirit
Avatar
Send message
Joined: 18 May 99
Posts: 6374
Credit: 28,630,854
RAC: 163
United States
Message 1072906 - Posted: 1 Feb 2011, 0:58:58 UTC

Due to planetary rotations and orbits, I would think no signal would be constant, but eventually repeating. So if we get 2 hits out of several hundred or thousand with a good gaussian, I think we got something of major interest.


____________

Janice

Profile William Rothamel
Avatar
Send message
Joined: 25 Oct 06
Posts: 2497
Credit: 1,174,617
RAC: 129
United States
Message 1073223 - Posted: 1 Feb 2011, 22:52:35 UTC - in response to Message 1072865.

Good questions--I would like a crisp definition of how we are revisiting promising results. I know that ARECIBO does not lend itself to doing this conveniently.

I am also curious why the Allen Array doesn't use BOINC--internacine warfare inhibiting our chances of success ??

You tell me.

Daddio

Profile tullioProject donor
Send message
Joined: 9 Apr 04
Posts: 3638
Credit: 366,799
RAC: 209
Italy
Message 1073278 - Posted: 2 Feb 2011, 2:08:16 UTC

From am article in "Nature" magazine of 16 September 2009 I have learned that the ATA has 3 users: the US Air Force, which covers its operating costs, the SETI Institute and the UCB Astronomy Dept. through its Radioastronomy Lab. The first in not interested in ETI, the second has its own project, Setiquest (see www.setiquest.org). Why the third is not cooperating with the Space Sciences Laboratory which manages SETI@home is a question I am not able to answer.
Tullio
____________

Message boards : SETI@home Science : I remember..

Copyright © 2014 University of California