As focusing on centre directions, rather than anti centre, is more scientifically profitable ( higher density of detectable pulsars that way ) then spend relatively more analysis and/or observation time in that direction. The PALFA observing time has to be slotted in with other jobs on the Aricebo site, and because there is an element of 'steerability' with reception it's not the case that everyone can get equal science value per antenna feed. With a given data set one can look for weaker signals, and other parameters like finer angular resolution etc, by grinding the handle of the maths machine harder or longer .....
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
As the telescope itself is stationary on the ground and can not move - only slightly shift the focus
Actually moving the receiver platform one can "shift the focus" quite a bit and track a given object or sky region for some time. But ultimately being bound to the earth's rotation the total observation time for a given direction is limited, and the time you are looking in one direction restricts the time available for other pointings.
The discussion on duration of anti-center pointings predates my involvement with P-ALFA (I think) but it has generally been true that they are sparser in pulsars (there's less stuff out there!) and that this affect survey strategy (on the other hand, it might be easier to get telescope time in those directions). I believe that the reason E@H didn't originally process them along with the others was something to do with the original template banks not working with shorter pointings (I think; Bernd could correct me on that matter). Anyhow, it's great that E@H is chewing through these as well.
Hmm. I see in the queue a new type of ABP jobs. Apparently they are made up of 10 micro-tasks, instead of the usual 4 (download 10 data files each, and CPU time x2.5). So it seems it still continued anti-center search only in large packages to reduce the load on the database?
BTW, all of my wingman has validation errors on them. (my not finish yet)
Hmm. I see in the queue a new type of ABP jobs. Apparently they are made up of 10 micro-tasks, instead of the usual 4 (download 10 data files each, and CPU time x2.5). So it seems it still continued anti-center search only in large packages to reduce the load on the database?
BTW, all of my wingman has validation errors on them. (my not finish yet)
Please see the home page news for info of the validation problems.
Indeed, bundling more work into a single unit should ease the load on the servers. It's also a preparation for the new revamped CUDA version (ABP3) that will be distributed in the (hopefully) not so distant future.
Now, i read news on home page about validations (note: my orig. post was 5 mins BEFORE post on homepage)
And what means is preparation? It is assumed that the new version will be much faster (due to transfer most of the calculations to the GPU) and, therefore, need to pack more data into a single WU to keep an optimal computation time?
P.S.
I think that ABP3 will be too late. By the time of its release is almost (or entirely) there will be no stock data from Arecibo. And current flow of new data does not require large computing resources ...
Much more interesting - is it possible to use CUDA (or much better OpenCL) in future for GW tasks? Or algoritm used there may not be mass parallelize?
I think that ABP3 will be too late. By the time of its release is almost (or entirely) there will be no stock data from Arecibo. And current flow of new data does not require large computing resources ...
There is an expectation of probably including data from radio telescopes other than Aricebo to include in the ABP work flow sometime 'soon'. The packing increase to 10 from 3 per WU is for distribution efficiency ( server loads, turnaround times .... ).
Quote:
..... may not be mass parallelize?
I think that's right. The GW side hasn't as many opportunities for that ....
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Do you think the data from other telescopes will (or can easily converted) to precisely the same format as it is now with Arecibo data? So that for their crunch can be used the same application?
Excellent if so :)
Anyway, using something else than the fasted available (and tested) code for the ABP search would be wasting volunteers' resources, so any improved version should be welcome as soon as it is stable.
I have no idea if data from other radio telescopes will be made available to the project, but for sure there is no technical reason why other telescopes' data streams would not be compatible with the ABP pipeline.
As for the GW search, as Bernd has pointed out in another thread, the GW (CPU only) app still has some potential for optimization (this is actually work in progress and looks promising). You want to know just how fast you can do it on the CPUs to have an idea whether you can do it significantly faster on the GPUs.
Data from the Allen Telescope Array go to the Radioastronomy Laboratory at Berkeley U. Data from Arecibo go to the Space Sciences Laboratory, also at Berkeley. Not being a Berkeley alumnus I do not know about their relationship.
Tullio
As focusing on centre
)
As focusing on centre directions, rather than anti centre, is more scientifically profitable ( higher density of detectable pulsars that way ) then spend relatively more analysis and/or observation time in that direction. The PALFA observing time has to be slotted in with other jobs on the Aricebo site, and because there is an element of 'steerability' with reception it's not the case that everyone can get equal science value per antenna feed. With a given data set one can look for weaker signals, and other parameters like finer angular resolution etc, by grinding the handle of the maths machine harder or longer .....
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: As the telescope itself
)
Actually moving the receiver platform one can "shift the focus" quite a bit and track a given object or sky region for some time. But ultimately being bound to the earth's rotation the total observation time for a given direction is limited, and the time you are looking in one direction restricts the time available for other pointings.
BM
BM
The discussion on duration of
)
The discussion on duration of anti-center pointings predates my involvement with P-ALFA (I think) but it has generally been true that they are sparser in pulsars (there's less stuff out there!) and that this affect survey strategy (on the other hand, it might be easier to get telescope time in those directions). I believe that the reason E@H didn't originally process them along with the others was something to do with the original template banks not working with shorter pointings (I think; Bernd could correct me on that matter). Anyhow, it's great that E@H is chewing through these as well.
Hmm. I see in the queue a new
)
Hmm. I see in the queue a new type of ABP jobs. Apparently they are made up of 10 micro-tasks, instead of the usual 4 (download 10 data files each, and CPU time x2.5). So it seems it still continued anti-center search only in large packages to reduce the load on the database?
BTW, all of my wingman has validation errors on them. (my not finish yet)
RE: Hmm. I see in the queue
)
Please see the home page news for info of the validation problems.
Indeed, bundling more work into a single unit should ease the load on the servers. It's also a preparation for the new revamped CUDA version (ABP3) that will be distributed in the (hopefully) not so distant future.
CU
HB
Now, i read news on home page
)
Now, i read news on home page about validations (note: my orig. post was 5 mins BEFORE post on homepage)
And what means is preparation? It is assumed that the new version will be much faster (due to transfer most of the calculations to the GPU) and, therefore, need to pack more data into a single WU to keep an optimal computation time?
P.S.
I think that ABP3 will be too late. By the time of its release is almost (or entirely) there will be no stock data from Arecibo. And current flow of new data does not require large computing resources ...
Much more interesting - is it possible to use CUDA (or much better OpenCL) in future for GW tasks? Or algoritm used there may not be mass parallelize?
RE: I think that ABP3 will
)
There is an expectation of probably including data from radio telescopes other than Aricebo to include in the ABP work flow sometime 'soon'. The packing increase to 10 from 3 per WU is for distribution efficiency ( server loads, turnaround times .... ).
I think that's right. The GW side hasn't as many opportunities for that ....
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Do you think the data from
)
Do you think the data from other telescopes will (or can easily converted) to precisely the same format as it is now with Arecibo data? So that for their crunch can be used the same application?
Excellent if so :)
Why would ABP3 be "too
)
Why would ABP3 be "too late"?
Anyway, using something else than the fasted available (and tested) code for the ABP search would be wasting volunteers' resources, so any improved version should be welcome as soon as it is stable.
I have no idea if data from other radio telescopes will be made available to the project, but for sure there is no technical reason why other telescopes' data streams would not be compatible with the ABP pipeline.
As for the GW search, as Bernd has pointed out in another thread, the GW (CPU only) app still has some potential for optimization (this is actually work in progress and looks promising). You want to know just how fast you can do it on the CPUs to have an idea whether you can do it significantly faster on the GPUs.
HB
Data from the Allen Telescope
)
Data from the Allen Telescope Array go to the Radioastronomy Laboratory at Berkeley U. Data from Arecibo go to the Space Sciences Laboratory, also at Berkeley. Not being a Berkeley alumnus I do not know about their relationship.
Tullio