2nd KRATOS COMMUNITY MEETING #8430
Replies: 7 comments 17 replies
-
|
Question proposed by @riccardotosi : Basilisk (http://basilisk.fr/) |
Beta Was this translation helpful? Give feedback.
-
|
@KratosMultiphysics/all Does anyone have experience with one of the open-source codes mentioned by @riccardotosi? If so, please let us know it by the end of the week (better by email at [email protected]). |
Beta Was this translation helpful? Give feedback.
-
|
Question proposed by @AlejandroCornejo : |
Beta Was this translation helpful? Give feedback.
-
|
I would go with the wheels for this rather than to come up with our own solution |
Beta Was this translation helpful? Give feedback.
-
|
Encapsulating in an executable binary is quite straightforward if we have a unified entry point, we are doing so. I think for a regular user that is not a programmer and just an end user would be the best option (and does not collide with the wheels option) |
Beta Was this translation helpful? Give feedback.
-
|
I have one doubt. Over the years, I have heard several times that Kratos can not directly compete in terms of speed with "x" commercial software as we are more focused on being very flexible. I understand this is due to some "core" decisions, so I'd like to know more exactly which those sacrifices are that were made in favor of flexibility and if they could be reverted/improved in the future. |
Beta Was this translation helpful? Give feedback.
-
|
Interesting, if x2-3 speed up in assembly can be achieved probably would be good to document the best practices. I'm pretty sure that we are overlooking other bottlenecks. Ideas on how to profile and spot them? Just to see what is achievable. |
Beta Was this translation helpful? Give feedback.
-
Blaze benchmarks: https://bitbucket.org/blaze-lib/blaze/wiki/Benchmarks (SPOILER: Blaze wins everything) |
Beta Was this translation helpful? Give feedback.
-
there is also a talk from the author of Blaze from the cppcon 2016. I am a fan of him, had some courses with him, he seems to really know what he is doing |
Beta Was this translation helpful? Give feedback.
-
|
Thanks to @RiccardoRossi for generating an extensive list of Kratos core bottlenecks coming from the flexibility!! And indeed, they are not few! Of course, finding the sweet spot in a trad-off situation like this is not very easy. I think we use to do it well and in many areas (like structural mechanics) we find a good compromise. But, every day we are opening to new solutions (like explicit ones) in which our balance is not valid anymore. I think maybe it is time to add a new way of adding physics to the Kratos (besides the current elements) which is more appropriate for the solutions that are less flexible by nature but more on the performance side. (Here I am thinking on DEM, explicit, and some fluids) For example to have the possibility of working with several same geometries together without the overhead of the dynamic allocations mentioned in many parts. Of course, here the problem is to make it compatible with the rest of the Kratos but I think it is doable. Now the question is if we have enough interest and resources to do it? |
Beta Was this translation helpful? Give feedback.
-
|
Apart from the very technical aspects listed, how does one properly compare for example Kratos FluidDyn and OpenFOAM CFD? As a user one is interested in the effort (human+machine) for a certain degree of quality of the results. Now omitting the human part, how do we compare it fairly? At least for me, even coming up with a fair (and usable) comparing scenario is hard. This only means in my view that one can and probably should focus on improvements in Kratos (which can be profiled/measured), whereas thinking about an objective (and useful) comparison with other software is very tricky. Or one can always plan the comparison such that the "winner" is known already when the "race" starts. |
Beta Was this translation helpful? Give feedback.
-
|
well, vtune is your friend...
…On Tue, Mar 9, 2021, 7:01 PM Jordi ***@***.***> wrote:
Interesting, if x2-3 speed up in assembly can be achieved probably would
be good to document the best practices. I'm pretty sure that we are
overlooking other bottlenecks. Ideas on how to profile and spot them? Just
to see what is achievable.
—
You are receiving this because you are on a team that was mentioned.
Reply to this email directly, view it on GitHub
<#8430 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AB5PWENF7WAL23QSJZZRDLDTCZO67ANCNFSM4YTUHQNA>
.
|
Beta Was this translation helpful? Give feedback.
-
|
And now is free 👯 |
Beta Was this translation helpful? Give feedback.
-
|
vtune and friend in the same sentence... vtune is the tool, but how to profile each of your proposals to know the impact is the hard part. |
Beta Was this translation helpful? Give feedback.
-
|
I think the (most) advantage of Kratos is its flexibility to combine
different components to make an analysis and we should really praise it. In
terms of efficiency, to profile and optimize a code can take more time than
developing a physics feature, and sometimes your solution is not the slow
path of the code but the correct solution strategy to converge. If you
design an analysis solution for a million engineer to use then it makes
sense to profile and optimize, writing code for a publication ... probably
not a good(best) idea.
On top of that, an idea for huge improvement in speed is to make a kind of
Compiler similar to Fenics (Dolfin). Based on the python code, the input,
confs, etc, generating an optimized executable using template-based
re-generation, vectorization, cache-aware, etc . That will be very useful
for those deploying the solution for industry problems where you need
speed. I hope this is not a stupid one :))
…On Tue, Mar 9, 2021 at 7:01 PM Jordi ***@***.***> wrote:
Interesting, if x2-3 speed up in assembly can be achieved probably would
be good to document the best practices. I'm pretty sure that we are
overlooking other bottlenecks. Ideas on how to profile and spot them? Just
to see what is achievable.
—
You are receiving this because you are on a team that was mentioned.
Reply to this email directly, view it on GitHub
<#8430 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ADJEUSQVX4IVWVTI65W6QJLTCZO67ANCNFSM4YTUHQNA>
.
|
Beta Was this translation helpful? Give feedback.
-
|
btw, i ordered today an epyc server with 24 cores to compile the wheels |
Beta Was this translation helpful? Give feedback.
-
|
Dear @KratosMultiphysics/all The recorded video of the 2nd KratosCommunityMeeting as well as the respective slides are available online. Topics addressed:
Video of the meeting Link to video |
Beta Was this translation helpful? Give feedback.
-
2nd KRATOS COMMUNITY MEETING
25th March 2021 at 9.00 (CET) @KratosMultiphysics/all
https://unipd.zoom.us/j/81286277636
This meeting follows the series started with the 1st Kratos Community Meeting
ADDRESSED TOPICS:
First, @philbucher will address the following technical topics:
Second, a question proposed by the community will be answered by a selected speaker. This question will be the most upvoted among those posted by the pasticipants in this discussion.
The meeting will last 30 minutes and we need YOUR HELP to organize it .
ACTION REQUIRED FROM YOUR SIDE:
• Post your questions, doubts, issues and suggestions as contributions to this discussion before 12th March 2021
• Use ‘thumb up’ or upvoting to support previously written questions. Some of the questions are ported from the previous meeting, because there was not enough time to address all of them. The most popular questions will have priority
AGENDA for 25th March 2021 at 9.00 (CET)
▪ Welcome, introduction and presentation of the selected questions (5 mins)
▪ CI, future of shared memory parallelization and C++ testing (10 mins)
▪ Second question is addressed (10 mins)
▪ Open discussion by ‘raising hands’ system (5 mins)
VIDEO AND MATERIAL OF THE MEETING
Video of the meeting Link to video
Pdf of the presentation by @philbucher Link to slides
@maceligueta @antonialarese @rolandwuechner
Beta Was this translation helpful? Give feedback.
All reactions