Gamma-ray pulsar search #5 (FGRP5)

Forums › Technical News

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4330
Credit: 251148968
RAC: 41551
9 Aug 2017 9:56:37 UTC
Topic 209161
(moderation:
    )

    We just started a new search for gamma-ray pulsars (CPU). This focuses on finding isolated pulsars from the galactic center. The application is actually the same as the current FGRPB1 1.08 (as is the FFT size), so you can use the same "wisdom" here (we didn't even change the name of the wisdom file).

    BM

    Bernd Machenschalk
    Bernd Machenschalk
    Moderator
    Administrator
    Joined: 15 Oct 04
    Posts: 4330
    Credit: 251148968
    RAC: 41551

    The CPU WUs of the search for

    9 Aug 2017 9:59:25 UTC
    Message 160780
    (moderation:
      )

      The CPU WUs of the search for gamma-ray pulsars in binary orbits (FGRPB1) will be phased out in favor of FGRP5.

      BM

      Gary Roberts
      Gary Roberts
      Moderator
      Joined: 9 Feb 05
      Posts: 5874
      Credit: 117980418230
      RAC: 21600386

      Bernd Machenschalk wrote:The

      10 Aug 2017 1:37:42 UTC
      Message 160801 in response to message 160780
      (moderation:
        )

        Bernd Machenschalk wrote:
        The CPU WUs of the search for gamma-ray pulsars in binary orbits (FGRPB1) will be phased out in favor of FGRP5.

        OK, sounds good but when is this actually going to happen.

        I had a look at the apps page and the new search #5 is listed with the V1.08 apps.  On the project preferences page, the old CPU search FGRPB1 is listed but not any new FGRP5 search.  I presume that tasks for this new search won't be distributed until it is listed and volunteers can re-jig their preferences as they see fit?

        At the moment I'm using a couple of locations (aka venues) to allow some hosts to use the V1.08-beta app whilst others are using the V1.05 previous app.  At a quick glance, both groups of hosts seem to be continuing to get their respective 'branded' tasks without issue.  As I saw a post mentioning that Bernd was on holidays, will the starting of the new search be delayed in any way due to this?

        Since the new FGRP5 V1.08 app is the same as the FGRPB1 V1.08-beta app, can the old app just be copied and renamed to the appropriate name and then pre-deployed to those hosts that will be doing FGRP5?  If so, could someone please advise the correct names for all apps?

        Will the procedure to create wisdom still require the manual running of the same special app?  Will its name still be the same?  Calculating wisdom on one machine and deploying to all others of the exact same CPU type seems to work fine, but I have a lot of different CPU types to potentially prepare for.  The number of different types for me would reduce quite a bit if the same wisdom could be used for members of a particular family rather than a particular model in that family. For example would the same wisdom work for all Haswell type CPUs?  What about the case where it's the same CPU but just a different frequency?

        Sorry to pester with a lot of questions.  I'm sure the answers might be useful to quite a few people.

         

        Cheers,
        Gary.

        Bernd Machenschalk
        Bernd Machenschalk
        Moderator
        Administrator
        Joined: 15 Oct 04
        Posts: 4330
        Credit: 251148968
        RAC: 41551

        Quote:Gary Roberts

        10 Aug 2017 5:20:00 UTC
        Message 160804 in response to message 160801
        (moderation:
          )

          Gary Roberts wrote:
          Bernd Machenschalk wrote:
          The CPU WUs of the search for gamma-ray pulsars in binary orbits (FGRPB1) will be phased out in favor of FGRP5.

          OK, sounds good but when is this actually going to happen.

          I had a look at the apps page and the new search #5 is listed with the V1.08 apps.  On the project preferences page, the old CPU search FGRPB1 is listed but not any new FGRP5 search.  I presume that tasks for this new search won't be distributed until it is listed and volunteers can re-jig their preferences as they see fit?

          We generated just a dozen test workunits yesterday just to test the pipeline. I guess we'll start at full speed after the weekend. Both searches will run in parallel for a while, until the last prepared FGRPB1 data sets are finished (~250k WUs).

          Quote:
          Will the procedure to create wisdom still require the manual running of the same special app?  Will its name still be the same?  Calculating wisdom on one machine and deploying to all others of the exact same CPU type seems to work fine, but I have a lot of different CPU types to potentially prepare for.  The number of different types for me would reduce quite a bit if the same wisdom could be used for members of a particular family rather than a particular model in that family. For example would the same wisdom work for all Haswell type CPUs?  What about the case where it's the same CPU but just a different frequency?

          You can use the exact same wisdom file for both applications on one machine, and even the name stays the same. If you haven't generated wisdom for FGRP5 yet, you do this in the exact same way under the same conditions (with the same tool) as described for FGRPB1 App v. 1.08.

          Wisdom generated with the tool will always "work", the question is how helpful it is in terms of how much it speeds up the application. The only answer I have about how similar a CPU needs to be to the one that the wisdom was generated on to get the same speedup and not to waste any potential to do better is: it depends. I really don't know the tiny details of what wisdom depends on (various frequencies, size of various caches etc.), and FFTW doesn't explain it in detail in the documentation. You may ask over at SETI, people there have been using wisdom for a long time.

          BM

          Bernd Machenschalk
          Bernd Machenschalk
          Moderator
          Administrator
          Joined: 15 Oct 04
          Posts: 4330
          Credit: 251148968
          RAC: 41551

          Gary Roberts wrote:Bernd

          10 Aug 2017 6:49:00 UTC
          Message 160808 in response to message 160801
          (moderation:
            )

            Gary Roberts wrote:
            Since the new FGRP5 V1.08 app is the same as the FGRPB1 V1.08-beta app, can the old app just be copied and renamed to the appropriate name and then pre-deployed to those hosts that will be doing FGRP5?  If so, could someone please advise the correct names for all apps?

            The application binaries are named 

            
hsgamma_FGRP5_1.08_i686-pc-linux-gnu__FGRPSSE

hsgamma_FGRP5_1.08_x86_64-apple-darwin__FGRPSSE

hsgamma_FGRP5_1.08_windows_intelx86__FGRPSSE.exe

hsgamma_FGRP5_1.08_x86_64-pc-linux-gnu__FGRPSSE

            The plan class is "FGRPSSE", beacuse these require SSE, it doesn't mean that they are limited to use that.

            BM

            Gary Roberts
            Gary Roberts
            Moderator
            Joined: 9 Feb 05
            Posts: 5874
            Credit: 117980418230
            RAC: 21600386

            Thank you very much.  That's

            10 Aug 2017 10:12:11 UTC
            Message 160810
            (moderation:
              )

              Thank you very much.  That's all very useful.  Much appreciated!!

              One final question, if I may.  Does 'wisdom' get updated at all as tasks are crunched, or, if a 'wisdom deficiency' is suspected, is the best course just to re-run the fftw app on that particular host to see if a better outcome can be achieved compared with the 'canned' one that might have been generated on a different host (same CPU)?  (I presume the latter but I thought I'd ask for complete clarity.)

              (It's a loooooonnng time since I've visited the Seti site, let alone asked a question there :-).  I don't know if I could cope with having to do that ;-).

               

              Cheers,
              Gary.

              Gary Roberts
              Gary Roberts
              Moderator
              Joined: 9 Feb 05
              Posts: 5874
              Credit: 117980418230
              RAC: 21600386

              Bernd Machenschalk wrote:You

              11 Aug 2017 21:57:00 UTC
              Message 160864 in response to message 160804
              (moderation:
                )

                Bernd Machenschalk wrote:
                You can use the exact same wisdom file for both applications on one machine, and even the name stays the same. If you haven't generated wisdom for FGRP5 yet, you do this in the exact same way under the same conditions (with the same tool) as described for FGRPB1 App v. 1.08.

                The new search is now listed on the project preferences page with all my 'locations' (aka venues) auto pre-selected - not quite what I wanted - so I'll have a job cleaning up the mess.  I have a venue for GPU only tasks with all CPU searches de-selected but the new search got auto pre-selected sometime overnight I guess.  That's OK, I'm not complaining, just pointing out that the auto opt-in for new searches causes issues for me when tasks become available very soon after the preferences are put up and I haven't had time to do something about it.  I have to sleep sometime :-).

                I just want to clarify one point about the name of the wisdom file.  You've said the name stays the same but I've noticed that the new app for generating wisdom is now called fftwf-wisdom_FGRP5_1.08_* compared with fftw-wisdom_FGRPB1_1.08_* when it was used with the beta tasks.  This is perfectly understandable so I'm just wondering if the FGRP5 V1.08 search app might just be looking for a wisdom file called FGRP5wisdom.dat rather than FGRPB1wisdom.dat.  I just want to be sure about that.

                 EDIT: Answered my own question.  I stopped BOINC on a GPU tasks only host that had inadvertently acquired a new FRGP5 CPU task and inserted a file with the old name FGRPB1wisdom.dat and then restarted BOINC.  I looked at the stderr.txt file in the appropriate slot directory and saw that the old name was recognised on startup.  Now I can go back and continue deploying the wisdom files, knowing for sure what the name should be :-).

                 

                Cheers,
                Gary.

                Gary Roberts
                Gary Roberts
                Moderator
                Joined: 9 Feb 05
                Posts: 5874
                Credit: 117980418230
                RAC: 21600386

                Having had the chance to go

                12 Aug 2017 5:20:06 UTC
                Message 160871
                (moderation:
                  )

                  Having had the chance to go through a lot of my hosts, seems like there may be just one that was given the new FGRP5 tasks.  I hadn't immediately looked at the server status page so didn't know the number was so limited.  Also I didn't notice initially, but the tasks I got (4 in total over a period) were resends of earlier failures. 

                  All the tasks so far issued seem to be failing (after crunching - not during) at the time when results are being uploaded.  Here is the error message at the end of the stderr.txt file.  It's the same -161 (not found) in a few I've looked at.

                  .....
% Following up candidate number: 10
% Refining in S
% Following-up in P
% C 17 11
% Writing follow-up output file.
FPU status flags:
18:58:49 (21216): [normal]: done. calling boinc_finish(0).
18:58:49 (21216): called boinc_finish
</stderr_txt>
<message>
upload failure: <file_xfer_error>
    <file_name>LATeah9999F_8.0_0_-2.9e-11_2_0</file_name>
    <error_code>-161 (not found)</error_code>
</file_xfer_error>
<file_xfer_error>
    <file_name>LATeah9999F_8.0_0_-2.9e-11_2_1</file_name>
    <error_code>-161 (not found)</error_code>
</file_xfer_error>

</message>
.....

                  Looks like some sort of problem with the file name :-(.

                   

                  Cheers,
                  Gary.

                  astro-marwil
                  astro-marwil
                  Joined: 28 May 05
                  Posts: 533
                  Credit: 657466543
                  RAC: 646721

                  Hallo! If it´s so easy to

                  12 Aug 2017 8:37:35 UTC
                  Message 160877
                  (moderation:
                    )

                    Hallo!

                    If it´s so easy to adapt to FGRP5, will there follow a version for GPU soon?

                    Kind regards and happy crunching.

                    Martin

                    Bernd Machenschalk
                    Bernd Machenschalk
                    Moderator
                    Administrator
                    Joined: 15 Oct 04
                    Posts: 4330
                    Credit: 251148968
                    RAC: 41551

                    Sorry, was offline over the

                    14 Aug 2017 12:44:00 UTC
                    Message 160946 in response to message 160871
                    (moderation:
                      )

                      "File not found": Sorry, was offline over the weekend. That's my fault, I used the wrong "result template". Fixed.

                      BM

                      Bernd Machenschalk
                      Bernd Machenschalk
                      Moderator
                      Administrator
                      Joined: 15 Oct 04
                      Posts: 4330
                      Credit: 251148968
                      RAC: 41551

                      astro-marwil wrote:If it´s so

                      23 Aug 2017 11:44:49 UTC
                      Message 161227 in response to message 160877
                      (moderation:
                        )

                        astro-marwil wrote:
                        If it´s so easy to adapt to FGRP5, will there follow a version for GPU soon?

                        The application code is the same, but the workunits are not. I think the current FGRP5 workunits transformed for GPUs would require 2GB GPU RAM minimum, in any case we'll need to set up a separate pipeline. So for now the GPUs will search for gamma-ray pulsars in binary systems.

                        BM

                        Comment viewing options

                        Select your preferred way to display the comments and click "Save settings" to activate your changes.

                        Forums › Technical News