arXiv subjects

Jihyun Kang

Publications and source records attributed to Jihyun Kang.

5 recordsLinked to original sources

arXiv2026

From Detection to Recovery: Operational Analysis on LLM Pre-training with 504 GPUs

Large-scale AI training is fundamentally a distributed systems problem, where hardware failures are routine operating conditions rather than rare exceptions, yet public operational evidence from production training clusters remains limited. This report presents an empirical analysis of a 63-node NVIDIA B200 production cluster (504 GPUs), using 55 days of Prometheus time-series data and 73 days of operational logs covering 224 multi-node training sessions. The environment is cross-organizational: five parties (SKT, Upstage, Lablup, NVIDIA Korea, VAST Data) share a unified monitoring pipeline. This enabled joint diagnosis of a 60-node-scale storage I/O bottleneck absent in 2-4-node tests, a production-scale phenomenon no single team could isolate alone. We perform three quantitative analyses yielding four findings. First, over 751 Prometheus metrics and 10 XID-identified GPU failures, no single metric is consistently dominant across failure types, motivating multi-signal detection. Second, 523 checkpoint events trace the save/load path from GPU VRAM to the NFS server: restart loading reaches 21.5% of maximum read bandwidth (700 GB/s) and save bursts 16.0% of maximum write bandwidth (250 GB/s), with NFS/RPC queueing and transport-layer backlog rising together. Third, across 224 sessions over 73 days, node exclusions concentrate so the top 3 of 63 nodes account for over 50%. Fourth, auto-retry chain analysis shows a 33.3% success rate over 12 chains (73 attempts), 2.7x the 12.5% manual rate, with a median retry interval of 11 minutes (IQR 10-11). All analyses are grounded in production infrastructure providing session-level workload management, GPU-centric scheduling, and unified observability.

cs.DC↗

arXiv2021

The JCMT BISTRO Survey: Evidence for Pinched Magnetic Fields in Quiescent Filaments of NGC 1333

We investigate the internal 3D magnetic structure of dense interstellar filaments within NGC 1333 using polarization data at $850 μ\mathrm{m}$ from the $B$-fields In STar-forming Region Observations survey at the James Clerk Maxwell Telescope. Theoretical models predict that the magnetic field lines in a filament will tend to be dragged radially inward (i.e., pinched) toward the central axis due to the filament's self-gravity. We study the cross-sectional profiles of the total intensity ($I$) and polarized intensity (PI) of dust emission in four segments of filaments unaffected by local star formation that are expected to retain a pristine magnetic field structure. We find that the filaments' FWHM in PI are not the same as those in $I$, with two segments being appreciably narrower in PI (FWHM ratio $\simeq 0.7-0.8$) and one segment being wider (FWHM ratio $\simeq 1.3$). The filament profiles of the polarization fraction ($P$) do not show a minimum at the spine of the filament, which is not in line with an anticorrelation between $P$ and $I$ normally seen in molecular clouds and protostellar cores. Dust grain alignment variation with density cannot reproduce the observed $P$ distribution. We demonstrate numerically that the $I$ and PI cross-sectional profiles of filaments in magnetohydrostatic equilibrium will have differing relative widths depending on the viewing angle. The observed variations of FWHM ratios in NGC 1333 are therefore consistent with models of pinched magnetic field structures inside filaments, and especially if they are magnetically near-critical or supercritical.

astro-ph.GA↗

arXiv2020

JCMT POL-2 and BISTRO Survey observations of magnetic fields in the L1689 molecular cloud

We present 850$μ$m polarization observations of the L1689 molecular cloud, part of the nearby Ophiuchus molecular cloud complex, taken with the POL-2 polarimeter on the James Clerk Maxwell Telescope (JCMT). We observe three regions of L1689: the clump L1689N which houses the IRAS 16293-2422 protostellar system, the starless clump SMM-16, and the starless core L1689B. We use the Davis-Chandrasekhar-Fermi method to estimate plane-of-sky field strengths of $366\pm 55$ $μ$G in L1689N, $284\pm 34$ $μ$G in SMM-16, and $72\pm 33$ $μ$G in L1689B, for our fiducial value of dust opacity. These values indicate that all three regions are likely to be magnetically trans-critical with sub-Alfvénic turbulence. In all three regions, the inferred mean magnetic field direction is approximately perpendicular to the local filament direction identified in $Herschel$ Space Telescope observations. The core-scale field morphologies for L1689N and L1689B are consistent with the cloud-scale field morphology measured by the $Planck$ Space Observatory, suggesting that material can flow freely from large to small scales for these sources. Based on these magnetic field measurements, we posit that accretion from the cloud onto L1689N and L1689B may be magnetically regulated. However, in SMM-16, the clump-scale field is nearly perpendicular to the field seen on cloud scales by $Planck$, suggesting that it may be unable to efficiently accrete further material from its surroundings.

astro-ph.GA↗

arXiv2020

Multiple outflows in the high-mass cluster forming region, G25.82-0.17

We present results of continuum and spectral line observations with ALMA and 22 GHz water (H$_2$O) maser observations using KaVA and VERA toward a high-mass star-forming region, G25.82-0.17. Multiple 1.3 mm continuum sources are revealed, indicating the presence of young stellar objects (YSOs) at different evolutionary stages, namely an ultra-compact HII region, G25.82-E, a high-mass young stellar object (HM-YSO), G25.82-W1, and starless cores, G25.82-W2 and G25.82-W3. Two SiO outflows, at N-S and SE-NW orientations, are identified. The CH$_3$OH 8$_{-1}$-7$_{0}$ E line, known to be a class I CH$_3$OH maser at 229 GHz is also detected showing a mixture of thermal and maser emission. Moreover, the H$_2$O masers are distributed in a region ~0.25" shifted from G25.82-W1. The CH$_3$OH 22$_{4}$-21$_{5}$ E line shows a compact ring-like structure at the position of G25.82-W1 with a velocity gradient, indicating a rotating disk or envelope. Assuming Keplerian rotation, the dynamical mass of G25.82-W1 is estimated to be $>$25 M$_{\odot}$ and the total mass of 20 M$_\odot$-84 M$_\odot$ is derived from the 1.3 mm continuum emission. The driving source of the N-S SiO outflow is G25.82-W1 while that of the SE-NW SiO outflow is uncertain. Detection of multiple high-mass starless$/$protostellar cores and candidates without low-mass cores implies that HM-YSOs could form in individual high-mass cores as predicted by the turbulent core accretion model. If this is the case, the high-mass star formation process in G25.82 would be consistent with a scaled-up version of low-mass star formation.

astro-ph.GA↗

arXiv2017

Inner warm disk of ESO H$α$ 279A revealed by NA I and CO overtone emission lines

We present analysis of near-infrared, high-resolution spectroscopy towards the Flat-spectrum YSO (Young Stellar Object) ESO H$α$ 279a (1.5 solar mass) in the Serpens star forming region, at the distance of 429 pc. Using the Immersion GRating INfrared Spectrometer (IGRINS, R=45,000), we detect emission lines originating from the accretion channel flow, jet, and inner disk. Specifically, we identify hydrogen Brackett series recombination, [Fe II], [Fe III], [Fe IV], Ca I, Na I, H2, H2O and CO overtone emission lines. By modeling five bands of CO overtone emission lines, and the symmetric double-peaked line profile for Na I emission lines, we find that ESO H$α$ 279a has an actively accreting Keplerian disk. From our Keplerian disk model, we find that Na I emission lines originate between 0.04 AU and 1.00 AU, while CO overtone emission lines are from the outer part of disk, in the range between 0.22 AU and 3.00 AU. It reveals that the neutral atomic Na gas is a good tracer of the innermost region of the actively accreting disk. We derive a mass accretion rate of 2-10x10^{-7} M_solar/yr from the measured Br_gamma emission luminosity of 1.78(+-0.31)x10^{31} erg/s.

astro-ph.SR↗