Re: [dev] Ada not Rust

From: Laslo Hunhold <dev_AT_frign.de>
Date: Wed, 12 May 2021 08:31:11 +0200

On Tue, 11 May 2021 11:15:37 +0300
Greg Minshall <minshall_AT_acm.org> wrote:

Dear Greg,

> i'm ignorant, but curious. a friend who does high performance
> computing is a fan of Julia, and in the past pointed me at
>
> https://www.nextplatform.com/2017/11/28/julia-language-delivers-petascale-hpc-performance/
>
> i'm not a believer in miracles, but the performance here seems fairly
> impressive. does this contradict your pessimistic assessment of Julia
> (without saying which view is correct)? is it maybe because of
> specialized features of Julia (i can imagine Julia's support for SIMD
> operations could help a lot with this sort of computing)? or...?
>
> obviously, incompatibility between versions, slow garbage collection,
> etc., which you mention below, are not good signs.

no language is perfect and Julia is still relatively young, but many
highly-respected institutions have been using it for years now,
including, but not limited to, BlackRock (for time-series analysis),
Federal Reserve NY (economy-modelling), Climate Modeling Alliance
(they're now using it exclusively for their climate models), NASA and
INPE (both for mission planning and orbital simulations) and many
researchers in physics and mathematics.

In terms of HPC, while low-level-performance to the metal will always
remain important, I know from experience that message-passing and
general parallelization are the key problems to solve for an algorithm
to run well, and Julia is really easy to use on clusters.

I'm just glad that I, as a numerical mathematician, don't have to use
MATLAB anymore. I initiated and finalized that the current lecture on
numerical mathematics here in Cologne, which I co-supervise, is using
Julia for the first time (instead of MATLAB), and I'm really happy
about that.

While C/Fortran are pure and great (and I've done my fair share of
HPC-programming in C with OpenMPI, BLAS, etc.), Julia is much more
suitable for the average Mathematician/Physicist who wants to implement
their algorithm. And given how well-readable it is, I find it
absolutely remarkable what performance it offers, despite some
drawbacks.

With best regards

Laslo
Received on Wed May 12 2021 - 08:31:11 CEST

This archive was generated by hypermail 2.3.0 : Wed May 12 2021 - 08:36:09 CEST