Message boards :
SETI@home Science :
Astropulse / Radar Blanking / Downtime
Message board moderation
Author | Message |
---|---|
Frizz Send message Joined: 17 May 99 Posts: 271 Credit: 5,852,934 RAC: 0 |
While we are all waiting for the new servers to come, would it be possible to have some thoughts on the way Astropulse handles radar blanking in the client? And if there would be a more efficient way? At the moment radar blanked data is replaced by random numbers with uniform distribution. This is done for each of the 888 dm_chunk_small loops. And this takes a very long time - especially for heavy blanked units, because a lot of double precision mul, div and log functions are used. (1) Would it be possible to use some sort of lookup table with, up to, state.datasize pre-generated random numbers? or, better: (2) "Do the radar removal exactly the same way they do it pre-split at the project. That is, after reading the raw data and figuring out which parts should be replaced, the replacement data overwrites those parts of the raw data. Subsequent processing treats all data as real. The advantage is that rather than doing replacement for each of the 888 dm_chunk_small loops it's only done once, and done linearly through the raw data rather than on 50% overlapped data chunks; it ought to reduce the time spent on removing radar by about a factor of 1600." (quoting Josef W. Segur / Lunatics board) Especially on the GPU ATI build runtime would be dramatically reduced. |
Eric Korpela Send message Joined: 3 Apr 99 Posts: 1382 Credit: 54,506,847 RAC: 60 |
Hi Frizz, I had exchanged a few messages with Josef Segur about this a while back. I was hoping that the use of the new software radar blanker would render the blanker in the client totally unnecessary. I haven't actually had time to figure out if it has. I'll have to take a look at how often the client blanker kicks in and what the average fraction blanked is on recent workunits. Eric @SETIEric@qoto.org (Mastodon) |
Frizz Send message Joined: 17 May 99 Posts: 271 Credit: 5,852,934 RAC: 0 |
Hello Eric, based on my observations of a few Hundred Workunits over the last 1/2 year I would say only about 20% are without blanking. The rest has low to medium blanking (some units almost 100% though :) . cheers, Frizz |
Frizz Send message Joined: 17 May 99 Posts: 271 Credit: 5,852,934 RAC: 0 |
After the outage, can we expect to get way less blanked units? Background: The software blanker is really starting to become showstopper for further accelerated GPU apps. |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
I would love to see this happen. Blanking is what currently slows down the ATI app. removing the data would certainly speed up the processing of work In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
musicplayer Send message Joined: 17 May 10 Posts: 2430 Credit: 926,046 RAC: 0 |
Doesn't (radar) blanking kill off the possible signal as well? |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
you understand that the signal would end up being buried under the local radar signal which is so strong that it would blot out anything worth looking at anyway. Blanking an area removes the offending signal. Yes we won't see ET if he's sending at that particular time and in that particular direction. In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Frizz Send message Joined: 17 May 99 Posts: 271 Credit: 5,852,934 RAC: 0 |
Doesn't (radar) blanking kill off the possible signal as well? Yes, it does. But that's not the "problem". The reason why I started this thread is: There are maybe *better* ways to do this blanking. Currently generating white noise, which replaces the RFI parts (-> blanks them), is very compute-intensive. Especially on the ATI GPU client units with blanking run 2x, 3x, ... times slower. It makes no sense IMHO that blanking takes more time than the actual data analysis. Petition against 1366x768 glare displays: http://www.facebook.com/home.php?sk=group_153240404724993 |
Frizz Send message Joined: 17 May 99 Posts: 271 Credit: 5,852,934 RAC: 0 |
I was hoping that the use of the new software radar blanker would render the blanker in the client totally unnecessary. ... After taking a look at my last couple of work units (200+) I still see lots of blanking. The three ones that are currently running on my system have 67%, 55% and 85% blanking. Question: How does this "software radar blanker" that you mentioned work? Since blanking is such a compute intensive task. E.g. for a heavy blanked tasks, the GPU Astropulse client spends about 3x as much time to do the blanking than on the actual signal analysis. How can/could this "software radar blanker" handle ALL units that are sent out to us crunchers at once? Petition against 1366x768 glare displays: http://www.facebook.com/home.php?sk=group_153240404724993 |
Silvester the furious Send message Joined: 19 Nov 10 Posts: 79 Credit: 1,734,928 RAC: 0 |
All this talk of blanking... Well, nevermind. "The hardest thing in the world to understand is the income tax." --Albert Einstein DON'T TREAD ON ME! |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.