Plans for near future of E@H ?

log in

Advanced search

Message boards : Science : Plans for near future of E@H ?

1 · 2 · 3 · 4 . . . 7 · Next
Author Message
astro-marwil
Send message
Joined: 28 May 05
Posts: 315
Credit: 43,554,661
RAC: 40,264
Message 119870 - Posted: 1 Nov 2012, 11:24:38 UTC

Hallo!
FGRP1 is out of data and S6LV1 will end close to the end of this year.
So I had a look into LSC-Virgo data analysis white paper - many thanks to joe areeda for posting this((:-))) -. But the only I found directly related to E@H was on page 131, in chaper 7 "LSC Computing and Software" the last block: "The LSC also developed the Einstein@Home project to leverage an alternative distributed computing paradigm for its most formidable computing challenge, the search for gravitational waves from isolated pulsars. The pulsar analysis puts reduced demand on quick turn-around and has low data flow, but requires PFlops of computing power. The ....". (We just reached 0.7PFLOPs.)
On A@H are no following up projects for test now.
For fresh data form the Advanced LIGO detectors we will have to wait for another year at minimum, according to the official time schedule.
So what will be the near future of E@H? Will we crunch BRP1 tasks on our CPUs, which is much less effective?

For a short answer we will be pleased.

Kind regards and happy crunching
Martin

P.S.: I found the "LSC-Virgo data analysis white paper" still very much interesting.
____________

Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 119874 - Posted: 1 Nov 2012, 14:57:38 UTC
Last modified: 1 Nov 2012, 15:12:40 UTC

For fresh data form the Advanced LIGO detectors we will have to wait for another year at minimum


Correct. But there may still something to be found in the existing data. For "continuous waves" searches the computing power is still what limits the sensitivity of a search, which roughly translates to the depth we could dig into space.

We are already preparing the next GW search, which will focus fewer promising sky positions, allowing us to reach out further than ever before.


Will we crunch BRP1 tasks on our CPUs, which is much less effective?


You think? In contrast to the GW search the Radio-Pulsar search lead to a couple of really interesting new discoveries. In my very personal view this is more effective than the GW searches we did so far.

We still haven't caught up with Arecibo data production, AFAIK we haven't yet processed any data from 2012. There are still a lot of Radio Pulsars out there waiting to be discovered!

And finally the Gamma-Ray Pulsar search is suspended because we ran out of manpower, not of data. I hope we can resume that work soon.


For a short answer we will be pleased.


Short answer to a long question?


BM
astro-marwil
Send message
Joined: 28 May 05
Posts: 315
Credit: 43,554,661
RAC: 40,264
Message 119875 - Posted: 1 Nov 2012, 16:19:18 UTC - in response to Message 119874.

Hallo BM!
Thank you for quick answering.

For "continuous waves" searches the computing power is still what limits the sensitivity of a search, which roughly translates to the depth we could dig into space.

I believe so. I remember, in early days of EaH we where happy reaching 100GFLOPS, and one task lasted much more than a year to finish. Now we have 7000 times of this crunching power, which gives more freedome for more complex project tasks and longer averaging times, resulting in higher sensitivity.

In contrast to the GW search the Radio-Pulsar search lead to a couple of really interesting new discoveries. In my very personal view this is more effective than the GW searches we did so far.

Sorry, my sentence was incomplete. I compared crunching on CPU / GPU. If I remmber correctly the BRP tasks where running for about 10h on CPU instead of some 10min on GPU. Comparing the requiered electrical power to crunch 1 BRP task, I suggest the GPU is also in forehand. Did someone test this? The figure is highly dependend on the used hardware.

We are already preparing the next GW search, which will focus fewer promising sky positions, allowing us to reach out further than ever before.

..... I hope we can resume that work soon.

We´re happy to heare this and do wait.

Kind regards and happy crunching
Martin
____________
Akos Fekete
Volunteer developer
Avatar
Send message
Joined: 13 Nov 05
Posts: 562
Credit: 4,410,312
RAC: 3
Message 119876 - Posted: 1 Nov 2012, 16:45:11 UTC - in response to Message 119874.

You think? In contrast to the GW search the Radio-Pulsar search lead to a couple of really interesting new discoveries. In my very personal view this is more effective than the GW searches we did so far.

So, Einstein@Home can profit from the results of pulsar hunting?
I always thought that this secondary project only diverts the resources.
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 119879 - Posted: 1 Nov 2012, 19:46:33 UTC - in response to Message 119876.
Last modified: 1 Nov 2012, 20:12:21 UTC

You think? In contrast to the GW search the Radio-Pulsar search lead to a couple of really interesting new discoveries. In my very personal view this is more effective than the GW searches we did so far.


So, Einstein@Home can profit from the results of pulsar hunting?
I always thought that this secondary project only diverts the resources.


The ultimate goal of Einstein@Home is to contribute to a better understanding of our universe by studying neutron stars, objects with such extreme conditions that they drive our theories of matter and forces to the limits (and possibly beyond), allowing to test and refine these.

The only way we can study these objects is through their emissions we could detect on (or near) earth - electromagnetic waves (like radio or gamma) or gravitational waves.

In that sense every result, in particular every new discovery of a pulsar, helps a bit to understand - what is possible, and what not, which theories and models do match best what we observe. So all three searches currently running on Einstein@Home do help the general goal.

As for the more narrow goal of detecting gravitational waves, the radio pulsar discoveries of Einstein@Home do help to form pulsar timing arrays that will help us to measure gravitational waves of a different wavelength than what the LIGO detectors and the GW search on Einstein@Home is sensitive for.

Finally the radio pulsar search allows us to keep and even increase our user base and thus computing power, even during periods when there is no new data from the GW detectors available - like now.

BM
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 119881 - Posted: 1 Nov 2012, 20:17:03 UTC - in response to Message 119875.
Last modified: 1 Nov 2012, 20:17:31 UTC

In contrast to the GW search the Radio-Pulsar search lead to a couple of really interesting new discoveries. In my very personal view this is more effective than the GW searches we did so far.

Sorry, my sentence was incomplete. I compared crunching on CPU / GPU. If I remmber correctly the BRP tasks where running for about 10h on CPU instead of some 10min on GPU. Comparing the requiered electrical power to crunch 1 BRP task, I suggest the GPU is also in forehand. Did someone test this? The figure is highly dependend on the used hardware.


So you meant 'efficient' instead of 'effective'.

BM
Profile Mike Hewson
Volunteer moderator
Avatar
Send message
Joined: 1 Dec 05
Posts: 4617
Credit: 39,756,848
RAC: 15,079
Message 119882 - Posted: 1 Nov 2012, 21:48:56 UTC
Last modified: 1 Nov 2012, 23:24:18 UTC

I remember the radio pulsar search as being initiated to solve several problems : an unavoidable hiatus in new data from the IFO's ie. keep the punters interested while we await GW detector upgrades, to help the PALFA consortium's need for processing their data, and to orthogonally study the same class of stars/systems as GW's will anyway. The gamma search is along the same lines.

If that understanding is correct then : the radio and gamma work have done brilliantly. As for whether it is 'correct' for the E@H project to have solved such issues is a higher question. In any case the value of GW work has not been lost, indeed upper limits have now been placed on certain signal types ( of interest to observers and theorists alike ) not to mention an enormous slab of valuable experience in running the analysis 'pipeline' that E@H is a part of. For the first, or for that matter subsequent, GW detections then validation and confidence in the entire enterprise will be crucial.

As a 'leading edge' activity I think it is quite reasonable & healthy to assess directions from time to time, to reflect upon progress or otherwise and try to make good guesses for the future. Else it wouldn't be called 'research' would it ?? :-)

Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 120271 - Posted: 21 Nov 2012, 10:08:56 UTC - in response to Message 119874.

We are already preparing the next GW search, which will focus fewer promising sky positions, allowing us to reach out further than ever before.


Actually we will first extend the current S6 LineVeto search a bit to higher frequencies.

A number of recent publications and own results of pulsar searches (in radio and gamma-ray) lead us to think that there is a larger population of 'younger' (i.e. faster spinning) pulsars than we expected when we initially set up the search.

Note that this is another example of interaction between the different pulsar searches on Einstein@Home.

BM
Profile Mike Hewson
Volunteer moderator
Avatar
Send message
Joined: 1 Dec 05
Posts: 4617
Credit: 39,756,848
RAC: 15,079
Message 120490 - Posted: 27 Nov 2012, 7:37:29 UTC - in response to Message 120271.

We are already preparing the next GW search, which will focus fewer promising sky positions, allowing us to reach out further than ever before.


Actually we will first extend the current S6 LineVeto search a bit to higher frequencies.

A number of recent publications and own results of pulsar searches (in radio and gamma-ray) lead us to think that there is a larger population of 'younger' (i.e. faster spinning) pulsars than we expected when we initially set up the search.

Note that this is another example of interaction between the different pulsar searches on Einstein@Home.

BM

Bernd what frequencies are we going out to ? Do we still have the WU count per frequency going like quadratic to frequency ?

Cheers, Mike.
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 120493 - Posted: 27 Nov 2012, 10:56:38 UTC - in response to Message 120490.
Last modified: 27 Nov 2012, 10:56:59 UTC

Bernd what frequencies are we going out to ?


This is still being discussed. Most likely an additional 50Hz or so. The higher we get, the more we would need to include a possible 2nd order frequency derivative. Our current code should be able to do that, but that hasn't been (sufficiently) validated yet.

Do we still have the WU count per frequency going like quadratic to frequency ?


Yes, as long as we stick to only first frequency derivative ("spindown").

We are currently working on the Application. It will be only slightly different, returning two result files, one containing the older "2F" statistics, the other containing the newer "LV". In principle we are extending both previous runs (S6LV1 and S6Bucket) at once.

BM
Profile Mike Hewson
Volunteer moderator
Avatar
Send message
Joined: 1 Dec 05
Posts: 4617
Credit: 39,756,848
RAC: 15,079
Message 120522 - Posted: 27 Nov 2012, 22:41:06 UTC

Ah I see. Not only faster spinning, but faster evolving too.

Cheers, Mike
____________
"I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

Filipe
Send message
Joined: 10 Mar 05
Posts: 124
Credit: 36,716,302
RAC: 34,719
Message 121658 - Posted: 28 Dec 2012, 14:36:40 UTC

We'll continue with the BRP4 search indefinitely. First we need to catch up with the data backlog of about 13,000 beams (this will take a couple of months at the current rates) then we'll continue in the steady-state. PALFA is taking about 50 new beams of data per day, on the average, so this search will be running for some years, at least for as long as PALFA is collecting new data.


Since we are processing an average of 150beams/day, does that mean that we are going to have a shortage of BRP4 tasks to feed all the GPU´s?
____________
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 121664 - Posted: 29 Dec 2012, 13:26:12 UTC - in response to Message 121658.
Last modified: 29 Dec 2012, 13:45:44 UTC

We'll continue with the BRP4 search indefinitely. First we need to catch up with the data backlog of about 13,000 beams (this will take a couple of months at the current rates) then we'll continue in the steady-state. PALFA is taking about 50 new beams of data per day, on the average, so this search will be running for some years, at least for as long as PALFA is collecting new data.


Since we are processing an average of 150beams/day, does that mean that we are going to have a shortage of BRP4 tasks to feed all the GPU´s?


Not necessarily.

- The 13,000 beams (now rather 10,000) refer to the data that has been so far copied to the AEI. There is a bit more data to fetch from Cornell (~6000 beams IIRC).

- We could widen the parameter space that the data is searched for. This will give us more workunits per data.

- There is an OpenCL version of the Gamma-Ray pulsar search being actively worked on, I think it should be ready before the BRP search catches up with the Arecibo data.

- There is also (yet) another attempt to make use of the GPUs for the GW search, but this will take a little longer. Certainly not for the planned "extension" run, but possibly for the next; if not ready for the run's start, then we still might issue it when the run's in progress.

- While at it, the GW "extension" run will be named "S6BucketLVE". It has been successfully tested over on Albert and the launch on Einstein is projected for Jan 9.

BM
Filipe
Send message
Joined: 10 Mar 05
Posts: 124
Credit: 36,716,302
RAC: 34,719
Message 121666 - Posted: 29 Dec 2012, 14:13:19 UTC

Thanks for the detalhed answer Bernd.

It's good to have a project who comunicate this well with the crunchers.

Filipe
____________

Filipe
Send message
Joined: 10 Mar 05
Posts: 124
Credit: 36,716,302
RAC: 34,719
Message 123188 - Posted: 2 Mar 2013, 12:17:58 UTC
Last modified: 2 Mar 2013, 12:18:09 UTC

- Are you catching up wit the the Arecibo data?

- Is there any news about the open cl app for the Gamma Ray shearch?
____________

Profile Bikeman (Heinz-Bernd Eggenstein)
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 28 Aug 06
Posts: 3420
Credit: 126,318,559
RAC: 175,669
Message 123193 - Posted: 2 Mar 2013, 16:21:46 UTC - in response to Message 123188.

- Are you catching up wit the the Arecibo data?

- Is there any news about the open cl app for the Gamma Ray shearch?


Hi!

As for the Gamma Ray search, I think we can expect to start a full test of an app on Albert@Home in the next 7- 10 days or so.

Cheers
HB

____________
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 123400 - Posted: 11 Mar 2013, 16:25:51 UTC - in response to Message 123188.

- Are you catching up wit the the Arecibo data?


The time until we catch up with the backlog of Arecibo data is roughly shown by the "Days to process remaining data" on the Server status page.

We're currently gathering more data to keep our analysis pipeline fed, e.g. from another survey of the Parkes telescope.

BM
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 123515 - Posted: 18 Mar 2013, 10:20:35 UTC - in response to Message 123193.

As for the Gamma Ray search, I think we can expect to start a full test of an app on Albert@Home in the next 7- 10 days or so.


FGRP2 OpneCL App launched on Albert.

BM
Sid
Send message
Joined: 17 Oct 10
Posts: 117
Credit: 125,358,619
RAC: 47,444
Message 123516 - Posted: 18 Mar 2013, 10:49:27 UTC - in response to Message 123515.



FGRP2 OpneCL App launched on Albert.

BM


Thank you to inform us. However I have a question.
Is this new application for ati only? Certainly, we'd like to have something similar for Nvidia.

Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar
Send message
Joined: 15 Oct 04
Posts: 3563
Credit: 115,040,747
RAC: 76,654
Message 123517 - Posted: 18 Mar 2013, 11:09:39 UTC

See the related thread on Albert@Home.

BM

1 · 2 · 3 · 4 . . . 7 · Next

Message boards : Science : Plans for near future of E@H ?


Home · Your account · Message boards

This material is based upon work supported by the National Science Foundation (NSF) under Grants PHY-1104902, PHY-1104617 and PHY-1105572 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

Copyright © 2016 Bruce Allen