Out-of-Core simulation systems often produce a massive amount of data that cannot t on the aggregate fast memory of the compute nodes, and they also require to read back these data for computation. As a result, I/O data movement can be a bottleneck in large-scale simulations. Advances in memory architecture have made it feasible and a ordable to integrate hierarchical storage media on large-scale systems, starting from the traditional Parallel File Systems (PFSs) to intermediate fast disk technologies (e.g., node-local and remote-shared NVMe and SSD-based Burst Bu ers) and up to CPU main memory and GPU High Bandwidth Memory (HBM). However, while adding additional and faster storage media increases I/O bandwidth, it pressures the CPU, as it becomes responsible for managing and moving data between these layers of storage. Simulation systems are thus vulnerable to being blocked by I/O operations. The Multilayer Bu er System (MLBS) proposed in this research demonstrates a general and versatile method for overlapping I/O with computation that helps to ameliorate the strain on the processors through asynchronous access. The main idea consists in decoupling I/O operations from computational phases using dedicated hardware resources to perform expensive context switches. MLBS monitors I/O tra c in each storage layer allowing fair utilization of shared resources. By continually prefetching up and down across all hardware layers of the memory and storage subsystems, MLBS transforms the original I/O-bound behavior of evaluated applications and shifts it closer to a memory-bound or compute-bound regime. The evaluation on the Cray XC40 Shaheen-2 supercomputer for a representative I/Obound application, seismic inversion, shows that MLBS outperforms state-of-the-art PFSs, i.e., Lustre, Data Elevator and DataWarp by 6.06X, 2.23X, and 1.90X, respectively. On the IBM-built Summit supercomputer, using 2048 compute nodes equipped with a total of 12288 GPUs, MLBS achieves up to 1.4X performance speedup compared to the reference PFS-based implementation. MLBS is also demonstrated on applications from cosmology, combustion, and a classic out-of-core computational physics and linear algebra routines.

京都大学 / 0048 / 新制・論文博士 / 博士(医学) / 乙第13156号 / 論医博第2143号 / 新制||医||1028(附属図書館) / 京都大学大学院医学研究科医学専攻 / (主査)教授 林 康紀, 教授 渡邉 大, 教授 影山 龍一郎 / 学位規則第4条第2項該当 / Doctor of Medical Science / Kyoto University / DFAM

Fast neural oscillations known as beta (12-30Hz) and gamma (30-100Hz) rhythms are recorded across several brain areas of various species. They have been linked to diverse functions like perception, attention, cognition, or interareal brain communication. The majority of the tasks performed by the brain involves communication between brain areas. To efficiently perform communication, mathematical models of brain activity require representing neural oscillations as sustained and coherent rhythms. However, some recordings show that fast oscillations are not sustained or coherent. Rather they are noisy and appear as short and random epochs of sustained activity called bursts. Therefore, modeling such noisy oscillations and investigating their ability to show interareal coherence and phase synchronization are important questions that need to be addressed. In this thesis, we propose theoretical models of noisy oscillations in the gamma and beta bands with the same properties as those observed in in \textit{vivo}. Such models should exhibit dynamic and statistical features of the data and support dynamic phase synchronization. We consider networks composed of excitatory and inhibitory populations. Noise is the result of the finite size effect of the system or the synaptic inputs. The associated dynamics of the Local Field Potentials (LFPs) are modeled as linear equations, sustained by additive and/or multiplicative noises. Such oscillatory LFPs are also known as noise-induced or quasi-cycles oscillations. The LFPs are better described using the envelope-phase representation. In this framework, a burst is defined as an epoch during which the envelope magnitude exceeds a given threshold. Fortunately, to the lowest order, the envelope dynamics are uncoupled from the phase dynamics for both additive and multiplicative noises. For additive noise, we derive the mean burst duration via a mean first passage time approach and uncover an optimal range of parameters for healthy rhythms. Multiplicative noise is shown theoretically to further synchronize neural activities and better explain pathologies with an excess of neural synchronization. We used the stochastic averaging method (SAM) as a theoretical tool to derive the envelope-phase equations. The SAM is extended to extract the envelope-phase equations of two coupled brain areas. The goal is to tackle the question of phase synchronization of noise-induced oscillations with application to interareal brain communication. The results show that noise and propagation delay are essential ingredients for dynamic phase synchronization of quasi-cycles. This suggests that the noisy oscillations recorded in \textit{vivo} and modeled here as quasi-cycles are good candidates for such neural communication. We further extend the use of the SAM to describe several coupled networks subject to white and colored noises across the Hopf bifurcation ie in both quasi-cycle and limit cycle regimes. This allows the description of multiple brain areas in the envelope-phase framework. The SAM constitutes an appropriate and flexible theoretical tool to describe a large class of stochastic oscillatory phenomena through the envelope-phase framework.

No description available.

Gamma-ray bursts (GRBs) are the most luminous phenomena in the Universe, explosions whoseenergy is generated by supernovae or mergers of dense objects such as neutron stars. The GRBemission is divided into the prompt emission phase characterized by γ-ray radiation and the afterglowof lower energy radiation. The prompt emission phase is still not understood; as of now, there aretwo leading descriptions: the photospheric- and the synchrotron models. The synchrotron model hashad great success in describing GRB spectra, and specifically some of the brightest ones, although notwithout issues such as some observations being at odds with theory. On the other hand, photosphericmodels have had problems too of how to broaden the spectrum in order to explain the observeddata. One explanation for this broadening is that Radiation Mediated Shocks (RMSs) dissipate energybelow the photosphere. In this report, a time resolved spectral analysis of the prompt emission of GRB160625B – a very bright GRB known to produce synchrotron-like emission – is done. Komrad is animplementation of the Kompaneets RMS Approximation (KRA), which is a dissipative photosphericmodel. Komrad is then used to fit a photospheric model to the prompt emission of GRB 160625Bin order to explore whether photospheric models can account for synchrotron-like emission spectra.Great statistical support is found for the photospheric model in comparison to standard GRB fittingfunctions as well as a synchrotron function which is indicative of the photospheric model being able toexplain a synchrotron-like spectra.

The aging population and increase of ambulatory surgeries have greatly increased strain on surgical and post-surgical staff that decreases the safety of care. Our overall goal is to find ways to decrease the time of anesthetic recovery to allow for more efficient post-surgical treatment. The specific aims of this study were to assess the correlations between neurocognitive recovery measures of attention and vigilance to brain dynamics. We analyzed reaction time via auditory psychomotor vigilance testing (aPVT) testing and the Richmond agitation-sedation scale (RASS) scores in 145 patients prior to and preceding surgery. Intraoperative electroencephalogram was also recorded for 115 of those patients. Data was analyzed to associate aPVT performance to recovery time and intraoperative brain dynamics. We found an association coefficient between reaction time and RASS recovery of 0.022 (p-value = 0.0001) showing a significant association. Further, we found age to be a significant confounding variable (p=0.04421) and included this in our association model. Lastly, there was no significant association found between intraoperative burst suppression and reaction time values (p=0.497). Overall, aPVT was found to be a robust test to assess recovery timeline in peri-operative anesthesia care unit patients. These results highlighted the potential use of an objective metric to track neurocognitive recovery after anesthesia, especially in elderly patients undergoing surgery.

A quantitative analysis was performed on a sample of Fast-Rise Exponential-Decay gamma-ray bursts using a continuous fitting function. The data were obtained from the Large Area detector catalogue of the Burst and Transient Source Experiment, which flew from 1991-2000 onboard the Compton Gamma-Ray Observatory. The purpose here is to tabulate from the fits quantities associated with gamma-ray bursts, emphasising peak intensity, duration, and characteristic rates of rising and falling. V/Vmax and duration analyses show that the sample of bursts is representative of the larger population of gamma-ray bursts. A modified asymmetric double sigmoidal was found to fit the background subtracted peaks for the majority of bursts. From the regressions the amplitude, duration and rising and falling characteristic times are defined for each burst, as well as a value describing the asymmetry of the peak. These values are compared with each other and with catalogued values of duration and V/Vmax.

No description available.

No description available.

No description available.