Search results
(1 - 3 of 3)
- Title
- A SPECTRAL ELEMENT SOLVER FOR SIMULATING TURBULENT FLOWS IN COMPLEX GEOMETRIES
- Creator
- Kandala, Sriharsha
- Date
- 2015, 2015-07
- Description
-
Flows in urban street canopies are quite complex both physically and geometrically and often unique to the specific canopy. Understanding the...
Show moreFlows in urban street canopies are quite complex both physically and geometrically and often unique to the specific canopy. Understanding the physics of these flows is important for various applications like prediction and control of dispersion in urban environments and efficient navigation of Micro-Air Vehicles (MAV) in gusty velocity fields among others. These flows are turbulent and the velocity fields are characterized by a wide range of spatial and temporal scales. Traditionally, given the exorbitant number of grid points required for accurate resolution of all flow features in computer simulations, experimental measurements supplemented with theory were the only feasible choice for understanding these flows. However with rapid increase in computing power and development of highly scalable algorithms to harness this power, numerical simulations are increasingly becoming feasible for higher Reynolds number flows. In the current work, flow in a model urban street canyon is studied using high-fidelity three-dimensional computational fluid dynamics simulations. Specsolve, a parallel spectral element solver capable of running parallel simulations utilizing thousands of processors, is developed for this purpose. The simulation domain used in this study consists of a 5 by 7 array of obstacles representative of a typical urban environment with the canyon aspect ratio corresponding to the skimming flow regime. These simulations do not use any turbulence model and are stabilized using a filtering procedure. Hot-wire data obtained from the wind tunnel experiments performed on an identical domain are used to prescribe realistic inflow boundary conditions upstream of the array. Numerical simulations were performed for cases where the flow is perpendicular to the array and with the flow at 15 degree angle of incidence. A grid resolution study is conducted to zero-in on the spectral element mesh required to resolve all important flow features for the 0 degree angle of incidence case. Mean velocity, coherent-structures and turbulence characteristics are used to describe the most important flow features in the domain. Streetwise evolution of flow is studied and the results indicate that flow reaches an equilibrium state by the third street.
Ph.D. in Mechanical and Aerospace Engineering, July 2015
Show less
- Title
- Beam Line Design for Fully Staged Two Beam Acceleration at the Argonne Wakefield Accelerator Facility
- Creator
- Neveu, Nicole
- Date
- 2018
- Description
-
Two beam acceleration (TBA) is a candidate for future high energy physics machines and FEL user facilities. This is a scheme in which an...
Show moreTwo beam acceleration (TBA) is a candidate for future high energy physics machines and FEL user facilities. This is a scheme in which an electron accelerator uses a ``drive'' beam to transport and supply the RF power needed for acceleration on a secondary and independent 'witness' accelerator. This technology is attractive for its potential to improve the efficiency and simplicity of large scale machines. At the Argonne Wakefield Accelerator Facility (AWA), research into this potential accelerator scheme is ongoing. Completed experiments include a simplified staging set up, where high-charge, 65 MeV drive bunch trains were injected from the RF photoinjector into decelerating structures to generate a few hundred MW's of RF power. This RF power was transferred through an RF waveguide to accelerating structures that were used to accelerate the witness beam. Staging refers to the sequential acceleration (energy gain) in two or more structures on the witness beam line. The main limitation in past experiments was difficulty achieving 100\% transmission in the second stage which resulted in lower power generation. AWA plans to demonstrate fully staged TBA, which requires a separate beam line for each decelerating/accelerating pair. In this thesis, design specifications and initial hardware tests needed for a new, independent beam line for TBA was done. Simulations of the drive line were done using the code OPAL. Since OPAL was new to the AWA group, a benchmark comparison with ASTRA and GPT was done to validate initial results. Then two optimization algorithms were investigated and used to optimize the drive line at 40 nC. Comparison of results between the two algorithms were done, with no major discrepancies found. Then large scale and parallel optimizations were done for the optics configuration in the fully staged TBA beam line design. A kicker was designed and incorporated into the drive beam line to accomplish a modular design so that each accelerating structure can be independently powered by a separate drive beam. Experimental measurements of the kicker indicate the angle increases linearly with the supplied voltage, and the angle achieved meets the design requirements for fully staged TBA. Optics optimization was done to minimize the beam size at the center of the decelerating structures to ensure good charge transmission. The resulting design will be the basis for proof of principle experiments that will take place at the AWA facility.
Show less
- Title
- Efficient and Practical Cluster Scheduling for High Performance Computing
- Creator
- Li, Boyang
- Date
- 2023
- Description
-
Cluster scheduling plays a crucial role in the high-performance computing (HPC) area. It is responsible for allocating resources and...
Show moreCluster scheduling plays a crucial role in the high-performance computing (HPC) area. It is responsible for allocating resources and determining the order in which jobs are executed. Existing HPC job schedulers typically leverage simpleheuristics to schedule jobs, but such scheduling policies struggle to keep pace with modern changes and technology trends. The study of this dissertation is motivated by two new trends in HPC community: the rapid growth of heterogeneous system infrastructure and the emergence of artificial intelligence (AI) technologies. First, existing scheduling policies are solely CPU-centric. In contrast, systems become more complex and heterogeneous, and emerging workloads have diverse resource requirements, such as CPU, burst buffer, power, network bandwidth, and so on. Second, previous heuristic scheduling approaches are manually designed. Such a manual design process prevents adaptive and informative scheduling decisions. A recent trend in HPC is to intertwine AI to better leverage the investment of supercomputers. This embrace of AI provides opportunities to design more intelligent scheduling methods. In this dissertation, we propose an efficient and practical cluster scheduling framework for HPC systems. Our framework leverages AI technologies and considers system heterogeneity. The framework comprises four major components. First, shared network systems such as dragonfly-based systems are vulnerable to performance variability due to network sharing. To mitigate workload interference on these shared network systems, we explore a dedicated scheduling policy. Next, emerging workloads in HPC have diverse resource requirements instead of being CPU-centric. To cater to this, we design an intelligent scheduling agent for multi-resource scheduling in HPC leveraging the advanced multi-objective reinforcement learning (MORL) algorithm. Subsequently, we address the issues with existing state encoding approaches in RL-driven scheduling, which either lack critical scheduling information or suffer from poor scalability. To this end, we present an efficient and scalable encoding model. Lastly, the lack of interpretability of RL methods poses a significant challenge to deploying RL-driven scheduling in production systems. In response, we provide a simple, deterministic, and easily understandable model for interpreting RL-driven scheduling. The proposed models and algorithms are evaluated with real job traces from production supercomputers. Experimental results show our schemes can effectively improve job scheduling in terms of both user satisfaction and system utilization.
Show less