A novel parallelisation paradigm of medium-fidelity modules in ASTRA

The ASTRA suite of codes [1] is a convenient framework for heat and transport mod-

elling of tokamak plasma. Despite being around for almost 50 years, the ASTRA code is

increasingly used in the tokamak community due to its interactive nature, the successful

coupling of several physics modules and equilibrium solvers, and its remarkable speed,

especially when it’s about stiff mdoels like TGLF [2] or QuaLiKiZ [3], thanks to the

Pereverzev-Corrigan algorithm [4].

The ASTRA code has been recently fully refurbished, moving to f90 and modules, and

a python pre-processor instead of the old fortran-based one [5]. A certain degree of

parallelisation has always been available in ASTRA, but a very low-level one, based on

Inter-Process Communication (IPC) and semaphores, involving a complicated sharing of

memory segments and several calls from Fortran to C and back, making this parallelisa-

tion very error prone, difficult to extend and not scalable.

The transport coefficients computed using theory-based models offer great margins of

parallelisation, because they can be calculated independently in each radial location.

In this paper we describe a novel MPI-based approach for parallelising the quasi-linear

models TGLF and QuaLiKiZ and the neoclassical code NEO [6]. We discuss the verifica-

tion of the MPI implementation by benchmarking the modelling of a reference case with

the prediction of the serial version of the codes and the IPC approach. A new scheme with

better separation of gradients calculation allows to parallelise the computation down to

single radial points, therefore allowing favourable scaling on clusters with several tenths

of CPU nodes. Depending on the amount of radial points and the number of CPU nodes,

this can lead to a gain in computing time by a factor ∼10, without any loss in accuracy

nor stability.

[1] G. V. Pereverzev, P. N. Yushmanov, An Automated System of TRansoprt Analysus, IPP

Report 5/42 (1991)

[2] G. M. Staebler et al., Plasma Phys. Control. Fusion 63 (2021) 01501

[3] C. Bourdelle et al., Phys. of Plasmas 14.11 (2007) 112501

[4] G. V. Pereverzev, G. Corrigan, Comput. Phys. Commun. 179(8) (2008) 579

[5] G. Tardini et al., in preparation for Nuclear Fusion

[6] E. Belli, J. Candy, Plasma Phys. Control. Fusion 50 (2008) 095010