Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Tamás Budai
    @budait
    and it has to support native execution. If we are using containers then those prerequisites are already in the image
    Javi Carnero
    @emepetres

    Question regarding Orchestration , to take smart decisions, you need information on application, expected memory usage and range of number cpu on which it could be executed no ?

    Yes among other things. The exactly metrics that the orchestrator will be using have yet to be decided, but cpu and memory use will be among them for sure.

    and it has to support native execution. If we are using containers then those prerequisites are already in the image

    In this case in the yml file we can put the dependencies to run the container, if needed :)

    Tamás Budai
    @budait
    ofc
    Guillaume Dollé
    @gdolle
    @victorsndvg What was the command you use to run paraview ?
    vlgrun ?
    Javi N.
    @fjaviernieto

    Question regarding Orchestration , to take smart decisions, you need information on application, expected memory usage and range of number cpu on which it could be executed no ?

    Yes among other things. The exactly metrics that the orchestrator will be using have yet to be decided, but cpu and memory use will be among them for sure.

    I would add to these ones... I think waiting time in HPC queues is also an interesting input, together with computation capability, measured in FLOPS/sec for example. Knowledge about the app modules will also support to understand which parts really need HPC resources.

    Keeping historical data and doing forecasts is something that we can include in the algorithm

    Tamás Budai
    @budait
    but at first we should define some ballpark figures for applications to start with. I think that MADF and Pilot developers (WP4-5) could send us some input regarding that
    I did something similar in the past (we had to predict how much time it takes to execute a matlab job) and there we introduced a metric that described the computational demand of that job (a simple integer). Of course in this case this is a bit more complicated
    Christophe Prud'homme
    @prudhomm
    @fjaviernieto @budait that is very interesting. as WP4 dev we could provide some insight for each madfs about the general rules to "configure"/"size" the system (what we do in our heads ;))
    Tamás Budai
    @budait
    :+1:
    Tamás Budai
    @budait
    @prudhomm is about to start the "HPC4EVERYONE" movement :D
    for students
    Guillaume Dollé
    @gdolle
    :D
    Javi N.
    @fjaviernieto
    With exascale, maybe that will be possible. They still don't have apps for using such huge computation power so... :P
    victorsndvg
    @victorsndvg
    yes @gdolle, vglrun command allows you to export opengl graphics through the remote desktop
    Regarding the pvserver question, I think you can use any, not reserved , port as output from a compute node. You can write to my colleagues of the communications department ( comunicaciones.cesga.es ), the can answer you more accurately. I will also ask about how to use pvserver from FT2
    Christophe Prud'homme
    @prudhomm
    @victorsndvg @gdolle it would be nice to have some doc about post-processing in the mso4sc book
    victorsndvg
    @victorsndvg
    Yes :+1:
    Christophe Prud'homme
    @prudhomm
    @gdolle perhaps you can spend some time to show @victorsndvg how that works ?
    Guillaume Dollé
    @gdolle
    @victorsndvg Yes I can show him during a break :+1:
    victorsndvg
    @victorsndvg
    :+1: thanks @gdolle
    Christophe Prud'homme
    @prudhomm
    @fjaviernieto perhaps we should ask for the slides and put them online ?
    or atleast in atos owncloud ?
    Tamás Budai
    @budait
    @prudhomm @atgeirr We had a discussion about logging and monitoring and I've put together a few slides about what options we are thinking about. I will share this document with all MADF/Pilot developers but for this time I just paste it here. Please take a look and share what you think about the options! https://docs.google.com/presentation/d/1slkfWluLotlixG7LcF-Mb3HxkHuzz2quv3qRjotpzCE/edit?usp=sharing
    Javi N.
    @fjaviernieto

    @fjaviernieto perhaps we should ask for the slides and put them online ?

    I think all of them are gathered in the presentation laptop, so I think we should have all of them. I'll ask Zoltán

    Atgeirr Flø Rasmussen
    @atgeirr
    @budait About logging: I think option B is the only doable way. Let's define some common "Events" (such as timestep complete, solver trouble, etc.) and for each simulator give patterns (regexp?) to extract data from existing log output.
    The possible inconsistency (developers change logging, break event patterns) should be mitigated by continuous integration testing.
    johanhoffman
    @johanhoffman
    @prudhomm is the slide stuck?
    Ok! :)
    Atgeirr Flø Rasmussen
    @atgeirr
    johanhoffman
    @johanhoffman
    @atgeirr :) Thanks!
    Christophe Prud'homme
    @prudhomm
    @atgeirr @johanhoffman and here is the paper on interlingua http://www.wired.co.uk/article/google-ai-language-create
    johanhoffman
    @johanhoffman
    @prudhomm Thanks!
    Christophe Prud'homme
    @prudhomm
    @budait @atgeirr I agree that option B might be the way to go
    Tamás Budai
    @budait
    Thanks for the feedback!
    Tamás Budai
    @budait
    @prudhomm great presentation. Question about the upcoming toolbox "Maxwell". Is it for Electromagnetic problems?
    Guillaume Dollé
    @gdolle
    @budait yes exaclty
    Tamás Budai
    @budait
    great, thanks :)
    Guillaume Dollé
    @gdolle
    :+1:
    Christophe Prud'homme
    @prudhomm
    @budait yes as @gdolle said
    @fjaviernieto they have a h2020 initiative to deploy the magnet technology to industry
    Tamás Budai
    @budait
    @all I've put together a few slides, because I had the impression that we have still some misconceptions on what docker can and cannot be used for (discussed this with @gdolle)
    Guillaume Dollé
    @gdolle
    :+1:
    victorsndvg
    @victorsndvg
    Great, very clear now, I think. Thanks @budait and @gdolle
    :+1:
    Christophe Prud'homme
    @prudhomm
    @budait :+1: thanks !
    Trophime
    @Trophime
    @budait @prudhomm I've managed to get Salome 8.2.0 working on Bash on Linux. So we can envision to have debs for Feel++ and HiFiMagnet to get a working "port" on Windows. The only tricky part is as usual the X11 server.
    I hope to be able to test Feelpp on Windows to see what are the perfs we can expect....
    Christophe Prud'homme
    @prudhomm
    @Trophime :+1: