This work was partially supported by the Russia Science Foundation (grant 18-11-00118).
References
1. Kasyanov V., Kasyanova E. Methods and system for cloud parallel programming // Proc. of the 21st Intern. Conf. on
Enterprise Information Systems, 2019, V. 1, P. 623-629.
2. Kasyanov V. N., Gordeev D. S., et al. The cloud parallel programming system CPPS: visualization and verification of
Cloud Sisal programs. Novosibirsk: NSU, 2020. (in Russian).
3. The web-site of the CPPS system. [Electron. resource]. URL: http://cpps.iis.nsk.su/description/ (the date of access:
24.05.2021).
A circular layout algorithm for attributed hierarchical graphs with ports
V. N. Kasyanov, A. M. Merculov, T. A. Zolotuhin
A. P. Ershov Institute of Informatics Systems SB RAS
Email: kvn@iis.nsk.su
DOI 10.24412/cl-35065-2021-1-02-26
Information visualization based on graph models is a key component of support tools for many applica-
tions in science and engineering [1]. The Visual Graph system [2] is intended for visualization of big amounts of
complex information on the basis of attributed hierarchical graph models [3]. In this paper, a circular layout
algorithm for attributed hierarchical graphs with ports and its effective implementation in the Visual Graph
system are presented.
This work was partially supported by the Russian Foundation for Basic Research (grant 18-07-00024).
References
1. Herman I., Melancon G., Marshall M. S., Graph visualization and navigation in information visualization: a survey //
IEEE Trans. on Visualization and Computer Graphics. 2000. V. 6, P. 24-43.
2. Kasyanov V. N., Zolotuhin T. A. A system for visualization of big attributed hierarchical graphs // Intern. J. of
Computer Networks & Communications. 2018. V.10, N.2. P. 55-67.
3. Kasyanov V. N. Kasyanova E. V. Information visualization based on graph models // Enterprise Information
Systems. 2013. V. 7, N. 2. P. 187-197.
NUMA-aware MPI broadcast algorithms for shared memory systems
M. G. Kurnosov1,2, E. I. Tokmasheva1
1Siberian State University of Telecommunications and Information
2Rzhanov Institute of Semiconductor Physics SB RAS
Email: mkurnosov@gmail.com
DOI 10.24412/cl-35065-2021-1-02-27
Broadcast is an important communication operation in high-performance computing. For a significant
number of parallel algorithms and packages of supercomputer simulation, the performance of broadcast op-
eration is critical. The MPI standard defines an MPI_Bcast routine for single source non-personalized broadcast
operation, in which data available at a root process is sent to all other processes. On HPC systems, MPI appli-
cations usually run several processes per compute node and therefore the latency of intra-node communica-
tions can significantly impact the performance of the overall application. Thus, optimization methods that lev-
erage intra-node shared memory become increasingly crucial. We consider a data broadcasting only between
processes reside on a same compute node. It is a typical situation for hierarchical (topology-aware) collective
operations to form a separate MPI communicators for each node and execute an algorithm of collective opera-
tion level-by-level.
The most used double copy algorithms (copy-in/copy-out) involve a shared buffer space used by local pro-
cesses to exchange messages. The root process copies the content of the message into the shared buffer be-
fore the receiver reads from it. In this paper we propose kernel-assisted (CMA, KNEM and XPMEM) and CICO-
based NUMA-aware algorithms for MPI_Bcast operation. In contrast to other works our algorithms explicitly
allocate memory for queues from local NUMA nodes even with active linux page cache readahead subsystem.
We show how to find optimal size f of buffer and length s of the queue what takes no more than b bytes and
provides minimum algorithm time. On NUMA machines with Xeon Nehalem and Xeon Broadwell processors,
our implementation based on Open MPI achieves on average 20�60% speedup over algorithms of Open MPI
coll/sm and MVAPICH (mv2_shm_bcast).
This work was (partially) supported by the Siberian Branch of Russian Academy of Science and the Russian Founda-
tion of Basic Research (grant 20-07-00039).
Aggregation errors in project schedulling
O. A. Lyakhov
Institute of Computational Mathematics and Mathematical Geophysics SB RAS
Email: loa@rav.sscc.ru
DOI 10.24412/cl-35065-2021-1-02-28
The problem of resources evaluation in network projects models with constant and variable intensity of
operations is discussed. The exactness of plans depends on a degree of resources aggregation and a time scale,
which consists of some intervals (quanta). A quantum is the least time period (called so by analogy with phys-
ics) for measuring resources and other purposes in scheduling. A sum of quanta is equal to planning period.
Resources demands for schedules are calculated separately per each quantum taken as a whole. A decrease of
a number of quanta (one way of aggregation) reduces models dimensions, but at the same time it leads to
growth of systematic mistakes of resources calculation, which doesn�t allow seeing schedules in proper per-
spective. The results of numerical experiments using the package for solving network optimization problems
[1] are presented.
This work was carried out under state contract with ICMMG SB RAS (0251-2021-0005).
References
1. Lyakhov O.A., Software package for network planning and scheduling. Novosibirsk, Foundation of algorithms and
programs, ICM&MG SB RAS, No. PR12002, 2012.
The diameter vulnerability of two-dimensional optimal circulant networks
E. A. Monakhova, O. G. Monakhov
Institute of Computational Mathematics and Mathematical Geophysics SB RAS
Email: monakhov@rav.sscc.ru
DOI 10.24412/cl-35065-2021-1-02-30
This paper studies the effect of changing the diameter of circulant networks of dimension two with unreli-
able elements. The well-known (Deg, D, D1, S)-problem is to find (Deg, D)-graphs with maximum degree Deg
and diameter D such that the subgraphs obtained from the original graph by deleting any set of up to S verti-
ces (edges) have diameter at most D1. For a family of optimal circulants of degree four we found the ranges of
orders of the graphs that preserve the diameter of the graph for one (two) vertex or edge failures. It is proved