116 제 2 발표장 (2 일금 ) 박근태 1, 최해천 1*, 최석호 2, 사용철 2, 권오경 3 ANALYSIS AND CONTROL OF TIP-LEAKAGE FLOW IN AN AXIAL FLOW FAN USING LARGE EDDY SIMULATION K. Park, H. Choi *, S. Choi, Y. Sa and O.K. Kwon 1., (tip-leakage flow, TLF).,., TLF,,. TLF casing treatment tip modification.,. (larege eddy simulation, LES) TLF,. 2., (R tip ) (U tip ) 547,000 0.036R tip. ( ) 0.346 (Fig. 2)., Navier-Stokes - LES. LES [1],,, [2]. 2. Fig. 3. 3, Fig. 1 Top and side views of the axial flow fan. 2.1,. - Fig. 1 Corresponding author E-mail: chio@snu.ac.kr Fig. 2 Total-to-static pressure rise coefficient (Ψ) and total-to-static efficiency (η) versus flow coefficient (Φ).
(2 일금 ) 제 2 발표장 117 Fig. 3 Schematic diagram of the coordinates, computational domain, and boundary conditions. TLF (tip-leakage vortex, TLV). (tip-separation vortex, TSV). TLV. TLV TSV. 2.3 TLV TSV casing treatment tip modification (Fig. 5). (groove) (stepped) (grooved) 14%. 3. Fig. 4 λ 2 iso-surfaces colored with static pressure coefficient with streamlines from an instantaneous flow field. TLF LES., TLF TLV, TLF TSV TSV TLV. 14%. Fig. 5 Schematic diagram of the casing treatments: (a) circumferential fence; (b) circumferential groove; (c) stepped blade tip; (d) grooved blade tip. 46, 200, 401. 200 (Fig. 2). KISTI Tachyon II (Intel Xeon X5570 2.93 GHz) 512 MPI. 2.2 Fig. 4 λ 2,.,,. (No. 20152020105600) (KSC-2017-G2-0010),. [1] 2010, Lee, J. et al., Dynamic global model for large eddy simulation of transient flow, Phys Fluids, Vol. 22, p. 075106. [2] 2006, Kim, D. and Choi, H., Immersed boundary method for flow around an arbitrarily moving body, J. Comput. Phys., Vol. 212, pp. 662~680.
118 제 2 발표장 (2 일금 )
(2 일금 ) 제 2 발표장 119
120 제 2 발표장 (2 일금 )
(2 일금 ) 제 2 발표장 121
122 제 2 발표장 (2 일금 )
(2 일금 ) 제 2 발표장 123
124 제 2 발표장 (2 일금 )
(2 일금 ) 제 2 발표장 125 김기하 1, 강지훈 2, 최정일 1* PARALLELIZATION OF AN INCOMPRESSIBLE THERMAL FLOW SOLVER USING MPI_ALLTOALL COMMUNICATION K.H Kim, J.H Kang and J.I Choi 1.. - (Rayleigh-Beńard convection). 2. Corresponding author E-mail: jic@yonsei.ac.kr 2.1 MPI Fully decoupled monolithic projection method(fdmpm) [3].,, Crank Nicolson 2 (central finite difference)., 2. Approximate factorization technique LU,,, projection.. approximate factorization Tridiagonal matrix algorithm(tdma). (Fourier diagonalization). FDMPM (iterative computation) TDMA fast Fourier transform(fft). FDMPM TDMA FFT Alltoall. Figure 1. Alltoall MPI (MPI communicator). Pack Unpack.
126 제 2 발표장 (2 일금 ) Alltoall Pack Unpack. (Derived data type). Pack Unpack Alltoall. 2.3 MPI MPI 1 3 ( ) 1 64. Figure 1 (scalability). Figure 1 (a). 16. Figure 1 (b).,. 2 Alltoall.. 3. -. Fully decoupled monolithic projection method(fdmpm). Alltoall MPI. MPI. Alltoall. non blocking Alltoall. Fig. 1 Scalability. (a) Scalability obtained for the total solver. (b) Scalability obtained for each major subroutine 본연구는한국과학기술정보연구원의지원을받아수행되 었습니다. [1] 1962, Kraichnan, Robert H, "Turbulent thermal convection at arbitrary Prandtl number", The Physics of Fluids,Vol.5 No.11, 1374-1389 [2] 2011, Grossmann, Siegfried and Lohse, Detlef, "Multiple scaling in the ultimate regime of thermal convection", Physics of fluids, Vol.23 No.4, 045108 [3] 2017, Pan, Xiaomin and Kim, Kyoungyoun and Lee, Changhoon and Choi, Jung-Il, "Fully decoupled monolithic projection method for natural convection problems", Journal of Computational Physics, Vol.334, 582-606
(2 일금 ) 제 2 발표장 127 김상혁 1, 권오경 2, 김종암 3* OPTIMIZATION ON HYBRID MPI-OPENMP PARALLEL PROGRAM FOR COMPRESSIBLE FLOW ANALYSIS BASED ON UNSTRUCTURED MIXED GRID S. Kim, O. Kwon and C. Kim 1. FVM(Finite Volume Method), ~. MPI. MPI, (Scalability)..,., OpenMP [1], MPI. MPI OpenMP (Hybrid), MPI [2]. OpenMP (race condition). MPI Corresponding author E-mail: chongam@snu.ac.kr OpenMP. 2. 2.1 (Race Condition) OpenMP, (write). (thread).. OpenMP. 2.1.1. ideal gas, 5 2 ( 1).,,,,,,. 5 (1)
128 제 2 발표장 (2 일금 ). 2 3,. OpenMP,. 2,..,. 2.1.2 Residual face flux cell residual. face 2 cell 1 residual face 3 flux cell 1 residual. - face (5) - cell (6) flux face ( 5), residual cell ( 6). 3. MPI OpenMP. (write),.. Fig. 1 cell face - face (2) (3) (4) face 2, face 2, cell 1, face 2, flux, cell 1, cell 1 residual, face 2. face, residual cell face. face face ( 2), cell residual ( 3, 4). (EDISON) (NRF-2011-0020576), KISTI (KSC-2018-C3-0009),. [1] 2011, Jung. Y. H., OpenMP Parallel Programming, freelec [2] 2014, Jeon. B.J. et al., Performance Analysis of the Parallel CUPID Code for Various Parallel Programming Models in Symmetric Multi-Processing System, Trans. Korean Soc. Mech. Eng. B, Vol. 38, No. 1, pp.71~79 [3] 2014, James Reinders, High Performance Parallelism Pearls Volume One: Multicore and Many-core Programming Approaches, Morgan Kaufmann
(2 일금 ) 제 2 발표장 129 강지훈 1* PERFORMANCE OPTIMIZATION AND LARGE-SCALE PARALLELIZATION OF TURBULENT FLOW SIMULATION USING KISTI-5 SUPERCOMPUTER Ji-Hoon Kang 1.,. Kim[1] (Direct Numerical Simulation) Re t =180 Re t =5000[2]. KISTI 4 Re t =3000 [3], KISTI 5. 4. KISTI 5 8,305... 2. 2.1 3 Navier-Stokes Crank-Nicolson fractional step.. 2.2 KISTI 5 5. MPI, OpenMP,. Corresponding author E-mail: jhkang@kisti.re.kr Fig.1 Simulation domain and boundary condition
130 제 2 발표장 (2 일금 ) 3. OpenMP MPI, 68 MPI OpenMP. OpenMP MPI, 4 8 4 OpenMP. Fig.2 Schematic of OpenMP and MPI parallelization MPI. x z 2, (non-blocking), (derived data-type),. OpenMP (directive). Fig. 2 MPI OpenMP.. (in-lining). 2.3 Fig. 3. 16. 16 weak scalability, 4 Fig.3 Scalability result 3. KISTI 5. MPI/OpenMP/vectorization, MPI OpenMP., 512 60.. (Intel Parallel Computing Center) (1711072904). [1] 1987, Kim, J. et al., Turbulence statistics in fully developed channel flow at low Reynolds number, J. Fluid Mech., Vol. 177, p. 133 [2] 2015, Lee, M. and Moser, R. D., Direct numerical simulation of turbulent channel flow up to Re τ 5200, J. Fluid Mech., Vol. 774, p. 295. [3] 2015, Ahn, J. et al., Direct numerical simulation of a 30R long turbulent pipe flow at Re τ 3008, Phys. Fluids, Vol. 27, 065110