Pith. sign in

REVIEW 3 major objections 4 minor 36 references

Load Balancing in Strongly Inhomogeneous Simulations -- a Vlasiator Case Study

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Using the global Vlasiator simulation as a testbed, the paper argues that Hilbert space-filling curve (HSFC) partitioning gives the best load balance among five tested methods, and that the hyperorthogonal Beta curve outperforms the…

desk verdict Solid engineering study with real data; the HSFC-over-other-algorithms ranking looks credible, but the Beta-curve claim is confounded by implementation details the authors themselves flag. read the letter →

arxiv 2505.20908 v1 pith:7J4OWL2A submitted 2025-05-27 cs.DC

Leo Kotipalo , Markus Battarbee , Yann Pfau-Kempf , Vertti Tarvus , Minna Palmroth This is my paper
classification cs.DC
keywords loadbalancingHilbertspace-fillingcurveVlasiatorhypergraphpartitioningrecursivecoordinatebisectioninertialLp-dilationadaptivemeshrefinement
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks which dynamic load-balancing heuristic best handles a simulation whose per-cell workloads vary by orders of magnitude, using the global Vlasov magnetosphere model Vlasiator as the test case. It compares five methods—hypergraph/graph partitioning (PHG), recursive coordinate bisection (RCB), recursive inertial bisection (RIB), and Hilbert space-filling curve (HSFC) partitioning—and, within HSFC, six three-dimensional Hilbert curves plus the discontinuous Z-curve, on small and production-scale runs. The central claim is that HSFC gives the best overall load balance, followed by RIB and PHG, with RCB last, and that the hyperorthogonal Beta curve beats the commonly used Butz/Octree curve by a few percent: about 4 percent in total propagation and 14 percent in spatial propagation on the larger run. The payoff is that production simulations with adaptive mesh refinement and sparse velocity space can gain speed without sacrificing accuracy, simply by choosing a curve with better locality. The paper also reports an unresolved discrepancy—functionally identical Butz and Octree implementations performed differently—which it attributes to coordinate ordering and flags as a priority for follow-up.

What carries the argument

The load-bearing object is the Hilbert space-filling curve used to map three-dimensional simulation cells onto a one-dimensional order, with load balancing performed by cutting the curve wherever cumulative cell weight reaches the target. The paper compares six self-similar 3D Hilbert curves from the same Ca00 base pattern, implemented as state tables for fast coordinate conversion; their quality is measured by $L_p$-dilation $W_{L_p}$, the maximum ratio of spatial distance to curve distance between two points, so low dilation means good locality. The Beta curve has the best $L_1$ and $L_2$ dilation among the six (cubic-root values $4.48$ and $2.65$, vs $4.62$ and $2.97$ for Butz/Octree), which is the property the paper credits for its faster spatial propagation. This machinery converts an NP-hard graph-partitioning problem into one-dimensional binning along a locality-preserving order, plus a local refinement step.

What would settle it

Run the production-scale workload with Beta and Butz implemented from identical state tables and identical coordinate ordering; if Beta no longer consistently beats Butz in spatial propagation across 12 trials, the claimed advantage of Beta over the standard curve is explained by coordinate ordering rather than by its lower $L_p$-dilation. Separately, instrument ghost-cell counts and communication volume per curve: the locality argument requires Beta's partitions to communicate less than Octree's, so observing no reduction would falsify the proposed mechanism.

Watch

Extended reading notes

Core claim

The paper's central discovery is that HSFC partitioning is the best load-balancing method among those compared for strongly inhomogeneous global simulations like Vlasiator: it matches the near-perfect weight balance of RCB and RIB (imbalance around $1\times10^{-3}$) while producing spatially compact partitions that lower communication, something the graph-based PHG methods achieve only with weight imbalances one to two orders of magnitude larger. Within the HSFC family, the paper finds the hyperorthogonal Beta curve—one of six 3D Hilbert curves sharing the base pattern Ca00, with the best L1 and L2 dilation among them—consistently outperforms the standard Butz/Octree curve: on the larger run the improvement is about 14 percent in spatial propagation and 4 percent in total propagation time. The interpretation is locality: Beta's lower $L_p$-dilation means points that are far apart on the curve are far apart in space, so greedy cuts along the curve yield connected, roughly cuboidal domains with fewer ghost cells. The paper is careful that the gain is not explained by L1 dilation alone, since Sasburg has lower L1 dilation than Butz but performed worse; Beta is better than Octree on all three dilation metrics considered.

Load-bearing premise

The curve comparison assumes the six Hilbert curve implementations differ only in their curve topology, yet the paper's own Octree-versus-Butz discrepancy shows implementation details such as coordinate ordering also change performance; if Beta's edge comes from those details rather than from its superior locality metrics, the main curve-related conclusion collapses.

Editorial extensions

If this is right

  • For Vlasiator's production-scale runs, HSFC with the Beta curve should be the default load-balancing choice: it improves spatial propagation by about 14 percent and total propagation by about 4 percent over the standard Octree/Butz curve.
  • Among the five algorithm families, the ranking is HSFC first, RIB and PHG (graph and hypergraph) next, and RCB last; the gap widens with run size.
  • Because HSFC balances cell weights almost as well as RIB/RCB while also optimizing communication, it avoids the large velocity-space and boundary-update penalties that hold graph/hypergraph partitioning back.
  • Switching Hilbert curves is a 'free' performance gain: it does not change simulation physics or accuracy, only the partition geometry.
  • The hyperorthogonal curves Alfa and Beta are identified as the expected optimal choices for load-balancing purposes, with Beta optimal for L1 and L2 dilation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the paper's own Octree-versus-Butz comparison shows that two implementations of the same curve perform differently, the reported gap between Beta and Octree may partly reflect coordinate ordering rather than the curve's intrinsic locality; a follow-up holding coordinate order fixed across all six curves would isolate the topology effect.
  • The dilation-based explanation predicts that Beta partitions should have measurably fewer ghost cells and lower communication volume than Octree partitions; instrumenting those quantities directly would test the mechanism more cleanly than wall-clock timers.
  • The same HSFC-with-Beta setup should transfer to any strongly inhomogeneous stencil or particle simulation whose communication follows spatial proximity, such as adaptive mesh refinement in astrophysics or combustion codes; the paper only demonstrates it for Vlasiator.
  • The results suggest that, in production, the overhead of building and maintaining a communication graph for graph partitioning may be unnecessary: a geometric method with a good curve achieves both balance and locality with simpler bookkeeping.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript reports an experimental comparison of load-balancing algorithms for the Vlasiator global magnetospheric simulation: hypergraph partitioning (PHG graph/hypergraph), recursive coordinate bisection (RCB), recursive inertial bisection (RIB), and Hilbert space-filling curves (HSFC), using two runs at different scales (S and L). The authors also compare seven curve implementations: Zoltan's Octree, a Lawder/King implementation of Butz, four further 3D Hilbert curves (Alfa, Harmonious, Sasburg, Base Camp, Beta), and the Z-curve. They conclude that HSFC provides the best load balance, followed by RIB and PHG, with RCB worst, and that the Beta Hilbert curve outperforms the commonly used Octree/Butz curve by a few percent, up to about 4% in total propagation time and 14% in spatial propagation on the larger run. The performance data come from wall-clock timers, 12 trials per configuration, and are complemented by memory usage and weight-imbalance measurements; the data and scripts are made available online.

Significance. If the central performance claims hold, the paper would be a valuable empirical contribution: it evaluates several 3D Hilbert curves from Haverkort's taxonomy in a real production-size simulation, provides a public dataset, and gives practitioners a concrete recommendation for a non-default curve in Zoltan-based HSFC partitioning. The use of two scales, 12 paired trials, and externally published locality metrics are strengths, as is the explicit falsification of the simple L1-dilation hypothesis via the Sasburg curve. However, the paper's main curve-level claim is currently not separable from implementation-dependent coordinate ordering, because two implementations of the same curve (Octree and Butz) differ in the reported measurements. The significance of the HSFC-vs-RIB algorithm-level conclusion is more robust, but it too is reported only through the Beta representative curve.

major comments (3)
  1. [Section III-B, V-B, VI] The central claim that the Beta curve intrinsically outperforms the standard Octree/Butz curve is confounded by implementation-dependent coordinate ordering. In Section III-B the authors state that they added a re-implementation of Butz with a different coordinate order alongside Zoltan's native Octree code, and Section V-B reports that Octree and Butz, which should be functionally identical, differ in both tests and attributes this discrepancy to coordinate order. Since Beta is implemented in the Lawder state-table formulation while Octree is Zoltan's native implementation, the reported 4% total and 14% spatial improvement on run L (Section V-B, Figure 13) cannot be separated from coordinate-order effects. Section VI explicitly lists testing different coordinate orderings as a priority, which confirms that this control is missing. I ask the authors to add controlled experiments, e.g., running Beta under Octree's coordinate ordering or permuting coordinate axes for each curve, and to report whether Beta's advantage survives; if it does not, the abstract and conclusion should be revised to claim only that their Beta implementation outperformed Zoltan's Octree implementation.
  2. [Section V, Figures 8, 10, 11, 13] The quantitative performance claims are reported without measures of uncertainty or significance tests. Twelve trials were run per configuration and the design is paired, since each job runs every method in succession, so paired comparisons or confidence intervals for the median or mean differences are straightforward to compute. This matters because the run-S curve comparison shows total propagation time that is "almost identical" and only about 1% spatial difference, and even the run-L Beta-vs-Octree margin should be reported with an interval to show it is not dominated by node-to-node variance. Similarly, Table III reports weight imbalance from a single run for each algorithm and cannot support strong statements about which algorithm has lower imbalance; if the imbalance is used as an explanatory variable, it should be measured over multiple runs or at least explicitly acknowledged as anecdotal.
  3. [Section V-A] The algorithm-level comparison uses Beta as the representative HSFC curve, where Beta was identified as best in the same experiment family (Section IV, Section V-B). This is not circular in the strict sense, since HSFC's advantage over RIB is large, but it makes the "HSFC optimal" claim contingent on the curve chosen. On run S, HSFC and RIB are almost identical in total propagation time, so if the Beta advantage over Octree is partly an implementation artifact, the algorithm-level ordering could change. Please also present or mention the algorithm comparison using Zoltan's default Octree curve, or state explicitly that the HSFC conclusion applies to the best-tested curve implementation rather than to HSFC with the standard curve.
minor comments (4)
  1. [Section V-B] The sentence "Octree and Butz curves, which should functionally identical" is missing the verb "be"; it should read "which should be functionally identical."
  2. [Table I] The table lists Octree and Butz as having identical dilations, while the experiments show they differ in runtime; a short footnote explaining that dilation is invariant under the coordinate-order change but the resulting partitions are not would help readers interpret the discrepancy.
  3. [Section IV] In the description of the two runs, "on each nodes" should be "on each node," and the phrase "the amount of cells per process" should be "the number of cells per process."
  4. [Section II-C] The definition of the greedy binning step says a cut is placed when the cumulative weight exceeds the target, but the later refinement step is only mentioned briefly; one sentence describing how cells are moved between bins and whether cuts are made exactly at target weights would clarify the partition construction.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the load-balancing comparisons are direct wall-time measurements, not derivations from fitted inputs.

full rationale

The paper's central claims—HSFC outperforms RIB/PHG/RCB and the Beta Hilbert curve outperforms the standard Octree/Butz curve—rest on direct Phiprof wall-time measurements across 12 replicated jobs, not on a parameter fitted to the outcome. The Hilbert locality metrics (L1/L2/L∞ dilation) are taken from Haverkort's external published enumeration, not from the authors' own prior work, so no ansatz is smuggled in via self-citation. The one potentially circular-looking sentence, 'using the best curve identified for HSFC which was Beta (Section V-B),' describes a normal two-stage experimental design: the curve screen informed which curve to use for the algorithm comparison; it does not define the algorithm comparison result in terms of itself. Self-citations [29]–[32] support simulation tools and data release, not the load-balancing conclusion. The Octree-versus-Butz discrepancy and the untested coordinate-ordering confound are threats to the internal validity of the curve-level comparison, but they are not circular reductions: no equation or fitted quantity is reused as its own prediction. Therefore no circular step is exhibited and the score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper makes no new theoretical claims and introduces no new entities. The main modeling inputs are the cell weight and communication cost assumptions, which are domain-specific but not fitted to the performance data. The Hilbert curves themselves are taken from published work.

free parameters (1)
  • Boundary cell weight factor = 1/6
    Chosen by hand to represent lower propagation cost of system boundary cells; applied uniformly to all methods, but affects the weight model and could influence balance quality.
assumptions (4)
  • domain assumption Block count is proportional to computational cost of a spatial cell
    Used to assign vertex weights in all partitioning methods (Section III-A). If this proportionality fails, all methods produce imbalanced partitions.
  • domain assumption Communication cost is proportional to velocity space block count of cells
    Used to weight edges in PHG and to reason about ghost cell cost for HSFC (Section III-A).
  • domain assumption Lower L1-dilation of a Hilbert curve leads to lower communication cost
    Hypothesis stated in Section II-D and revisited in Section VI; not fully confirmed by the data since Sasburg has lower L1 dilation than Octree but performs worse.
  • standard math Haverkort's enumeration of Hilbert curves is correct
    Relies on Haverkort for curve properties and dilation values (Table I).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Load Balancing in Strongly Inhomogeneous Simulations -- a Vlasiator Case Study." pith.science (2026). https://pith.science/paper/7J4OWL2A

@misc{pith2026250520908,
  author       = {Pith},
  title        = {Pith review of: Load Balancing in Strongly Inhomogeneous Simulations -- a Vlasiator Case Study},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7J4OWL2A}},
  note         = {Machine review of arXiv:2505.20908}
}
read the original abstract

Parallelization is a necessity for large-scale simulations due to the amount of data processed. In this article we investigate different load balancing methods using Vlasiator, a global magnetospheric simulation as our case study. The theoretical basis for load balancing is the (hyper)graph partitioning problem, modeling simulation units as vertices and their data dependencies as edges. As it is an NP-hard problem, heuristics are necessary for dynamic runtime balancing. We consider first hypergraph partitioning via an algorithm called parallel hypergraph partitioner (PHG); this is done by partitioning a simplified grid and then attempting to optimize the solution on the finer grid. The second and third are the geometric methods of recursive coordinate bisection (RCB) and recursive inertial bisection (RIB). Finally we consider the method of Hilbert space filling curves (HSFC). The algorithm projects simulation cells along a Hilbert curve and makes cuts along the curve. This works well due to the excellent locality of Hilbert curves, and can be optimized further by choice of curve. We introduce and investigate six three-dimensional Hilbert curves in total. Our findings on runs of two different scales indicate the HSFC method provides optimal load balance, followed by RIB and PHG methods and finally by RCB. Of the Hilbert curves evaluated, the Beta curve outperformed the most commonly used curve by a few percent.

Figures

Figures reproduced from arXiv: 2505.20908 by the authors.

Figure 1
Figure 1. Load balancing weights (color on a logarithmic scale) in a three-dimensional Vlasiator run, two slices with grid (black cubic mesh) overlaid. As can [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of graph partitioning. The objective is to partition the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Examples of PHG graph (a) and hypergraph (b) partitioning, one slice of a three-dimensional simulation with [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Examples of recursive coordinate (a) and inertial bisection (b), one slice of a three-dimensional simulation with [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The third order approximation of the two-dimensional Z-curve (a) and the first three orders of approximation of the two-dimensional Hilbert curve [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Example of Hilbert space filling curve partitioning using Zoltan’s [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: The six three-dimensional Hilbert curves evaluated: Butz (a), Alfa (b), Harmonious (c), Sasburg (d), Base Camp (e) and Beta (f). The grid here is a [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Box plots of core-seconds spent in total propagation (a), spatial propagation (b), velocity space propagation (c) and system boundary updates (d) for [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Box plots of resident memory per node at the end of simulation for the algorithms tested on run S (a) and run L (b), 12 trials. Box plot parameters [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Box plots of core-seconds spent in total propagation (a), spatial propagation (b), velocity space propagation (c) and system boundary updates (d) for [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Box plots of core-seconds spent in total propagation (a), spatial propagation (b), velocity space propagation (c) and system boundary updates (d) for [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Box plot of resident memory for the space filling curves tested on runs S (a) and L (b), 12 trials. The poor locality of the Base camp and Z-curves [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: Box plot of core-seconds spent in total propagation (a), spatial propagation (b), velocity space propagation (c) and system boundary updates (d) for [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 31 canonical work pages

  1. [1]

    Vlasiator: First global hybrid-Vlasov simulations of Earth’s foreshock and magnetosheath,

    S. von Alfthan, D. Pokhotelov, Y . Kempf, S. Hoilijoki, I. Honkonen, A. Sandroos, and M. Palmroth, “Vlasiator: First global hybrid-Vlasov simulations of Earth’s foreshock and magnetosheath,” Journal of Atmospheric and Solar-Terrestrial Physics , vol. 120, pp. 24–35, 2014. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1364682614001916

    work page 2014

  2. [2]

    fmihpc/vlasiator: Vlasiator 5.3.1,

    Y . Pfau-Kempf, U. Ganse, M. Battarbee, L. Kotipalo, T. Koskela, S. von Alfthan, I. Honkonen, M. Alho, A. Sandroos, K. Papadakis, H. Zhou, M. Palmu, M. Grandin, J. Suni, V . Tarvus, D. Pokhotelov, J. Hokkanen, V . Haaja, A. Menendez Pratt, K. Horaites, and A. Lalag ¨ue, “fmihpc/vlasiator: Vlasiator 5.3.1,” Sep. 2024. [Online]. Available: https://doi.org/1...

    work page doi:10.5281/zenodo.13752779 2024

  3. [3]

    Zoltan data management service for parallel dynamic applications,

    K. Devine, E. Boman, R. Heapby, B. Hendrickson, and C. Vaughan, “Zoltan data management service for parallel dynamic applications,” Computing in Science and Engg. , vol. 4, no. 2, p. 90–97, Mar. 2002. [Online]. Available: https://doi.org/10.1109/5992.988653

    work page arXiv 2002

  4. [4]

    Reducibility among combinatorial problems,

    R. M. Karp, “Reducibility among combinatorial problems,” in Complexity of Computer Computations: Proceedings of a symposium on the Complexity of Computer Computations, held March 20–22, 1972, at the IBM Thomas J. Watson Research Center, Yorktown Heights, New York, and sponsored by the Office of Naval Research, Mathematics Program, IBM World Trade Corporat...

    1972

  5. [5]

    D. S. J. Michael R. Garey, Computers and Intractability: A guide to the Theory of NP-Completeness. Bell Telephone Laboratories Incorporated, 1979, A5.2 Multiprocessor scheduling

    work page 1979

  6. [6]

    Bounds on multiprocessing timing anomalies,

    R. L. Graham, “Bounds on multiprocessing timing anomalies,” SIAM Journal on Applied Mathematics , vol. 17, no. 2, pp. 416–429, 1969, theorems 2 and 3. [Online]. Available: https://doi.org/10.1137/0117039

    doi:10.1137/0117039 1969

  7. [7]

    Graph partitioning and constructing optimal decision trees are polynomial complete problems,

    L. Hyafil and R. L. Rivest, “Graph partitioning and constructing optimal decision trees are polynomial complete problems,” IRIA – laboratoire de recherche en informatique et automatique, Tech. Rep. 33, Oct. 1973

    work page 1973

  8. [8]

    Berge, Graphs and Hypergraphs , ser

    C. Berge, Graphs and Hypergraphs , ser. North-Holland Mathematical Library. Elsevier, 1973, vol. 6, Chapter 17 – Hypergraphs and their Duals

    work page 1973

Show all 36 references
  1. [9]

    Hypergraph-partitioning-based decom- position for parallel sparse-matrix vector multiplication,

    U. Catalyurek and C. Aykanat, “Hypergraph-partitioning-based decom- position for parallel sparse-matrix vector multiplication,” IEEE Transac- tions on Parallel and Distributed Systems , vol. 10, no. 7, pp. 673–693, Jul. 1999

    1999

  2. [10]

    Fast balanced partitioning is hard even on grids and trees,

    A. E. Feldmann, “Fast balanced partitioning is hard even on grids and trees,” Theoretical Computer Science , vol. 485, pp. 61–68, 2013. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S0304397513002259

    2013

  3. [11]

    PARMETIS: Parallel graph partitioning and sparse matrix ordering library,

    G. Karypis, K. Schloegel, and V . Kumar, “PARMETIS: Parallel graph partitioning and sparse matrix ordering library,” University of Minnesota, Department of Computer Science, Tech. Rep. 97-060, 7a

  4. [12]

    PT-Scotch: A tool for efficient parallel graph ordering,

    C. Chevalier and F. Pellegrini, “PT-Scotch: A tool for efficient parallel graph ordering,” Parallel Computing , vol. 34, no. 6, pp. 318–331, 2008, parallel Matrix Algorithms and Applications. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0167819107001342

    2008

  5. [13]

    Boman, K

    E. Boman, K. Devine, L. A. Fisk, R. Heaphy, B. Hendrickson, C. Vaughan, U. Catalyurek, D. Bozdag, W. Mitchell, and J. Teresco, Zoltan 3.0: Parallel Partitioning, Load-balancing, and Data Management Services; Developer’s Guide , Sandia National Laboratories, Albuquerque, NM, 20...

    2007

  6. [14]

    A refinement-tree based partitioning method for dynamic load balancing with adaptively refined grids,

    W. F. Mitchell, “A refinement-tree based partitioning method for dynamic load balancing with adaptively refined grids,” Journal of Parallel and Distributed Computing, vol. 67, no. 4, pp. 417–429, 2007. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S07...

    2007

  7. [15]

    CPU nodes – Lumi-C,

    “CPU nodes – Lumi-C,” Feb. 2024. [Online]. Available: https: //docs.lumi-supercomputer.eu/hardware/lumic/

    2024

  8. [16]

    Parallel hypergraph partitioning for scientific computing,

    K. Devine, E. Boman, R. Heaphy, R. Bisseling, and U. Catalyurek, “Parallel hypergraph partitioning for scientific computing,” in Proceedings 20th IEEE International Parallel & Distributed Processing Symposium , 2006, pp. 10 pp.–. [Online]. Available: https://doi.org/10.1109/IP...

    2006 arXiv

  9. [17]

    Performance of dynamic load balancing algorithms for unstructured mesh calculations,

    R. D. Williams, “Performance of dynamic load balancing algorithms for unstructured mesh calculations,” Concurrency: Practice and Experience, vol. 3, no. 5, pp. 457–481, 1991. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/cpe.4330030502

    1991 doi

  10. [18]

    A partitioning strategy for nonuniform problems on multiprocessors,

    Berger and Bokhari, “A partitioning strategy for nonuniform problems on multiprocessors,” IEEE Transactions on Computers, vol. C-36, no. 5, pp. 570–580, 1987

    1987

  11. [19]

    A computer oriented geodetic data base; and a new technique in file sequencing,

    G. Morton, “A computer oriented geodetic data base; and a new technique in file sequencing,” International Business Machines Co. Ltd., Ottawa, Canada, Tech. Rep., Mar

  12. [20]

    ¨Uber die stetige Abbildung einer Linie auf ein Fl¨achenst¨uck,

    D. Hilbert, “ ¨Uber die stetige Abbildung einer Linie auf ein Fl¨achenst¨uck,” Mathematische Annalen , vol. 38, no. 3, pp. 459–460, Sep. 1891. [Online]. Available: https://doi.org/10.1007/BF01199431

    doi

  13. [21]

    How many three-dimensional Hilbert curves are there?

    H. Haverkort, “How many three-dimensional Hilbert curves are there?” Journal of Computational Geometry , vol. 8, no. 1, pp. 206–281, Sep

  14. [22]

    Alternative algorithm for Hilbert’s space-filling curve,

    A. Butz, “Alternative algorithm for Hilbert’s space-filling curve,” IEEE Transactions on Computers , vol. C-20, no. 4, pp. 424–426, 1971. [Online]. Available: https://doi.org/10.1109/T-C.1971.223258

    1971

  15. [23]

    Using state diagrams for Hilbert curve mappings,

    J. Lawder and P. King, “Using state diagrams for Hilbert curve mappings,” International Journal of Computer Mathematics , vol. 78, no. 3, pp. 327–342, 2001. [Online]. Available: https://doi.org/10.1080/ 00207160108805115

    2001

  16. [24]

    POV-Ray, the persistence of vision raytracer,

    “POV-Ray, the persistence of vision raytracer,” Nov. 2013. [Online]. Available: https://www.povray.org/

    2013

  17. [25]

    Enabling technology for global 3D + 3V hybrid-Vlasov simulations of near-Earth space,

    U. Ganse, T. Koskela, M. Battarbee, Y . Pfau-Kempf, K. Papadakis, M. Alho, M. Bussov, G. Cozzani, M. Dubart, H. George, E. Gordeev, M. Grandin, K. Horaites, J. Suni, V . Tarvus, F. T. Kebede, L. Turc, H. Zhou, and M. Palmroth, “Enabling technology for global 3D + 3V hybrid-Vla...

    2023 doi

  18. [26]

    Vlasov methods in space physics and astrophysics,

    M. Palmroth, U. Ganse, Y . Pfau-Kempf, M. Battarbee, L. Turc, T. Brito, M. Grandin, S. Hoilijoki, A. Sandroos, and S. von Alfthan, “Vlasov methods in space physics and astrophysics,” Living Reviews in Computational Astrophysics , vol. 4, no. 1, p. 1, Aug. 2018. [Online]. Avail...

    2018 doi

  19. [27]

    A three-dimensional monotone and conservative semi-lagrangian scheme (SLICE-3D) for transport problems,

    M. Zerroukat and T. Allen, “A three-dimensional monotone and conservative semi-lagrangian scheme (SLICE-3D) for transport problems,” Quarterly Journal of the Royal Meteorological Society , vol. 138, no. 667, pp. 1640–1651, 2012. [Online]. Available: https://rmets.onlinelibrary...

    2012 doi

  20. [28]

    Spatial filtering in a 6D hybrid-Vlasov scheme to alleviate adaptive mesh refinement artifacts: a case study with Vlasiator (versions 5.0, 5.1, and 5.2.1),

    K. Papadakis, Y . Pfau-Kempf, U. Ganse, M. Battarbee, M. Alho, M. Grandin, M. Dubart, L. Turc, H. Zhou, K. Horaites, I. Zaitsev, G. Cozzani, M. Bussov, E. Gordeev, F. Tesema, H. George, J. Suni, V . Tarvus, and M. Palmroth, “Spatial filtering in a 6D hybrid-Vlasov scheme to al...

    2022

  21. [29]

    Physics- motivated cell-octree adaptive mesh refinement in the Vlasiator 5.3 global hybrid-Vlasov code,

    L. Kotipalo, M. Battarbee, Y . Pfau-Kempf, and M. Palmroth, “Physics- motivated cell-octree adaptive mesh refinement in the Vlasiator 5.3 global hybrid-Vlasov code,” Geoscientific Model Development , vol. 17, no. 16, pp. 6401–6413, 2024. [Online]. Available: https: //gmd.coper...

    2024

  22. [30]

    Phiprof – Parallel hierarchical profiler,

    S. von Alfthan, M. Battarbee, L. Kotipalo, Y . Pfau-Kempf, K. Papadakis, and I. Honkonen, “Phiprof – Parallel hierarchical profiler,” 2023. [Online]. Available: https://github.com/fmihpc/phiprof

    2023

  23. [31]

    Load balancing test configuration and data,

    L. Kotipalo and V . Tarvus, “Load balancing test configuration and data,” https://doi.org/10.23729/fd-9e84a409-0da9-3223-ad8b-011fd6112ce3, May 2025, University of Helsinki, Matemaattis-luonnontieteellinen tiedekunta

    2025 doi

  24. [32]

    Analysator,

    M. Battarbee, M. Alho, Y . Pfau-Kempf, U. Ganse, M. Grandin, L. Kotipalo, K. Papadakis, R. J ¨arvinen, L. P ¨ank¨al¨ainen, J. Suni, S. von Alfthan, V . Tarvus, I. Zaitsev, S. Tao, K. Horaites, L. Turc, F. K. Tesema, H. Zhou, I. Honkonen, T. Brito, A. Lalag ¨ue, S. Siljamo, and...

    2024

  25. [33]

    Technical details about Roihu,

    “Technical details about Roihu,” Mar. 2025. [Online]. Available: https://docs.csc.fi/computing/systems-roihu/

    2025

  26. [1966]

    Available: https://dominoweb.draco.res.ibm.com/ 0dabf9473b9c86d48525779800566a39.html

    [Online]. Available: https://dominoweb.draco.res.ibm.com/ 0dabf9473b9c86d48525779800566a39.html

  27. [1997]

    Available: https://hdl.handle.net/11299/215345

    [Online]. Available: https://hdl.handle.net/11299/215345

  28. [2017]

    Available: https://doi.org/10.20382/jocg.v8i1a10

    [Online]. Available: https://doi.org/10.20382/jocg.v8i1a10

    doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.