According to the data affinity, DAFA re-organizes data to maximize the parallelism of the affinitive data, and also subjective to the overall load balance. This enables DAFA to realize the maximum number of map tasks with data-locality. Besides the system performance, power consumption is another important concern of current computer systems. In the U.S. alone, the energy used by servers which could be saved comes to 3.17 million tons of carbon dioxide, or 580,678 cars {Kar09}. However, the goals of high performance and low energy consumption are at odds with each other. An ideal power management strategy should be able to dynamically respond to the change (either linear or nonlinear, or non-model) of workloads and system configuration without violating the performance requirement. We propose a novel power management scheme called MAR (modeless, adaptive, rule-based) in multiprocessor systems to minimize the CPU power consumption under performance constraints. By using richer feedback factors, e.g. the I/O wait, MAR is able to accurately describe the relationships among core frequencies, performance and power consumption. We adopt a modeless control model to reduce the complexity of system modeling. MAR is designed for CMP (Chip Multi Processor) systems by employing multi-input/multi-output (MIMO) theory and per-core level DVFS (Dynamic Voltage and Frequency Scaling).; TRAID deduplicates this overlap by only logging one compact version (XOR results) of recovery references for the updating data. It minimizes the amount of log content as well as the log flushing overhead, thereby boosts the overall transaction processing performance. At the same time, TRAID guarantees comparable RAID reliability, the same recovery correctness and ACID semantics of traditional transactional processing systems. On the other hand, the emerging myriad data intensive applications place a demand for high-performance computing resources with massive storage. Academia and industry pioneers have been developing big data parallel computing frameworks and large-scale distributed file systems (DFS) widely used to facilitate the high-performance runs of data-intensive applications, such as bio-informatics {Sch09}, astronomy {RSG10}, and high-energy physics {LGC06}. Our recent work {SMW10} reported that data distribution in DFS can significantly affect the efficiency of data processing and hence the overall application performance. This is especially true for those with sophisticated access patterns. For example, Yahoo's Hadoop {refg} clusters employs a random data placement strategy for load balance and simplicity {reff}. This allows the MapReduce {DG08} programs to access all the data (without or not distinguishing interest locality) at full parallelism. Our work focuses on Hadoop systems. We observed that the data distribution is one of the most important factors that affect the parallel programming performance. However, the default Hadoop adopts random data distribution strategy, which does not consider the data semantics, specifically, data affinity. We propose a Data-Affinity-Aware (DAFA) data placement scheme to address the above problem. DAFA builds a history data access graph to exploit the data affinity.; The evolution of computer science and engineering is always motivated by the requirements for better performance, power efficiency, security, user interface (UI), etc {CM02}. The first two factors are potential tradeoffs: better performance usually requires better hardware, e.g., the CPUs with larger number of transistors, the disks with higher rotation speed; however, the increasing number of transistors on the single die or chip reveals super-linear growth in CPU power consumption {FAA08a}, and the change in disk rotation speed has a quadratic effect on disk power consumption {GSK03}. We propose three new systematic approaches as shown in Figure 1.1, Transactional RAID, data-affinity-aware data placement DAFA and Modeless power management, to tackle the performance problem in Database systems, large scale clusters or cloud platforms, and the power management problem in Chip Multi Processors, respectively. The first design, Transactional RAID (TRAID), is motivated by the fact that in recent years, more storage system applications have employed transaction processing techniques Figure 1.1 Research Work Overview] to ensure data integrity and consistency. In transaction processing systems(TPS), log is a kind of redundancy to ensure transaction ACID (atomicity, consistency, isolation, durability) properties and data recoverability. Furthermore, high reliable storage systems, such as redundant array of inexpensive disks (RAID), are widely used as the underlying storage system for Databases to guarantee system reliability and availability with high I/O performance. However, the Databases and storage systems tend to implement their independent fault tolerant mechanisms {GR93, Tho05} from their own perspectives and thereby leading to potential high overhead. We observe the overlapped redundancies between the TPS and RAID systems, and propose a novel reliable storage architecture called Transactional RAID (TRAID).

This dissertation discusses the development of an architecture and associated techniques to support Privacy Preserving and Distributed Data Mining. The field of Distributed Data Mining (DDM) attempts to solve the challenges inherent in coordinating data mining tasks with databases that are geographically distributed, through the application of parallel algorithms and grid computing concepts. The closely related field of Privacy Preserving Data Mining (PPDM) adds the dimension of privacy to the problem, trying to find ways that organizations can collaborate to mine their databases collectively, while at the same time preserving the privacy of their records. Developing data mining algorithms for DDM and PPDM environments can be difficult and there is little software to support it. In addition, because these tasks can be computationally demanding, taking hours of even days to complete data mining tasks, organizations should be able to take advantage of high-performance and parallel computing to accelerate these tasks. Unfortunately there is no such framework that is able to provide all of these services easily for a developer. In this dissertation such a framework is developed to support the creation and execution of DDM and PPDM applications, called APHID (Architecture for Private, High-performance Integrated Data mining). The architecture allows users to flexibly and seamlessly integrate cluster and grid resources into their DDM and PPDM applications. The architecture is scalable, and is split into highly de-coupled services to ensure flexibility and extensibility. This dissertation first develops a comprehensive example algorithm, a privacy-preserving Probabilistic Neural Network (PNN), which serves a basis for analysis of the difficulties of DDM/PPDM development. The privacy-preserving PNN is the first such PNN in the literature, and provides not only a practical algorithm ready for use in privacy-preserving applications, but also a template for other data intensive algorithms, and a starting point for analyzing APHID's architectural needs. After analyzing the difficulties in the PNN algorithm's development, as well as the shortcomings of researched systems, this dissertation presents the first concrete programming model joining high performance computing resources with a privacy preserving data mining process. Unlike many of the existing PPDM development models, the platform of services is language independent, allowing layers and algorithms to be implemented in popular languages (Java, C++, Python, etc.). An implementation of a PPDM algorithm is developed in Java utilizing the new framework. Performance results are presented, showing that APHID can enable highly simplified PPDM development while speeding up resource intensive parts of the algorithm.

Traditional HPC scientific workloads are tightly coupled, while emerging scientific workflows exhibit even more complex patterns, consisting of multiple characteristically different stages that may be IO-intensive, compute-intensive, or memory-intensive. New high-performance computer systems are evolving to adapt to these new requirements and are motivated by the need for performance and efficiency in resource usage. On the other hand, cloud workloads are loosely coupled, and their systems have matured technologies under different constraints from HPC. In this thesis, the use of cloud technologies designed for loosely coupled dynamic and elastic workloads is explored, repurposed, and examined in the landscape of HPC in three major parts. The first part deals with the deployment of HPC workloads in cloud-native environments through the use of containers and analyses the feasibility and trade-offs of elastic scaling. The second part relates to the use of workflow management systems in HPC workflows; in particular, a molecular docking workflow executed through Airflow is discussed. Finally, object storage systems, a cost-effective and scalable solution widely used in the cloud, and their usage in HPC applications through MPI I/O are discussed in the third part of this thesis. / Framväxande vetenskapliga applikationer är mycket datatunga och starkt kopplade. Nya högpresterande datorsystem anpassar sig till dessa nya krav och motiveras av behovet av prestanda och effektivitet i resursanvändningen. Å andra sidan är moln-applikationer löst kopplade och deras system har mogna teknologier som utvecklats under andra begränsningar än HPC. I den här avhandlingen diskuteras användningen av moln-teknologier som har mognat under löst kopplade applikationer i HPC-landskapet i tre huvuddelar. Den första delen handlar om implementeringen av HPC-applikationer i molnmiljöer genom användning av containrar och analyserar genomförbarheten och avvägningarna av elastisk skalning. Den andra delen handlar om användningen av arbetsflödeshanteringsystem i HPC-arbetsflöden; särskilt diskuteras ett molekylär dockningsarbetsflöde som utförs genom Airflow. Objektlagringssystem och deras användning inom HPC, tillsammans med ett gränssnitt mellan S3-standard och MPI I/O, diskuteras i den tredje delen av denna avhandling / <p>QC 20231122</p>

High performance computing is a topic that has risen to the top in the era ofdigitalization, AI and automation. Therefore, the search for more cost and timeeffective ways to implement HPC work is always a subject extensively researched.One part of this is to have hardware that is capable to improve on these criteria. Different hardware usually have different code languages to implement theseworks though, cross-platform solution like Intel’s oneAPI framework is startingto gaining popularity.In this thesis, the capabilities of Intel’s oneAPI framework to implement andexecute HPC benchmarks on different hardware platforms will be discussed. Using the hardware available through Intel’s DevCloud services, Intel’s Xeon Gold6128, Intel’s UHD Graphics P630 and the Arria10 FPGA board were all chosento use for implementation. The benchmarks that were chosen to be used wereGEMM (General Matrix Multiplication) and BUDE (Bristol University DockingEngine). They were implemented using DPC++ (Data Parallel C++), Intel’s ownSYCL-based C++ extension. The benchmarks were also tried to be improved uponwith HPC speed-up methods like loop unrolling and some hardware manipulation.The performance for CPU and GPU were recorded and compared, as the FPGAimplementation could not be preformed because of technical difficulties. Theresults are good comparison to related work, but did not improve much uponthem. This because the hardware used is quite weak compared to industry standard. Though further research on the topic would be interesting, to compare aworking FPGA implementation to the other results and results from other studies. This implementation also probably has the biggest improvement potential,so to see how good one could make it would be interesting. Also, testing someother more complex benchmarks could be interesting.

High-performance Computing (HPC) systems play pivotal roles in societal and scientific advancements, executing up to quintillions of calculations every second. As we shift towards exascale computing and beyond, modern HPC systems emphasize resource sharing, where various applications share processors, memory, networks, and other components. While this sharing enhances power efficiency, it complicates performance prediction and introduces significant variations in application running times, affecting overall system efficiency and operational costs. HPC systems utilize monitoring frameworks that gather numerical telemetry data on resource usage to track operational status. Given the massive complexity and volume of this data, manual analysis is often daunting and inefficient. Machine learning (ML) techniques offer automated performance anomaly diagnosis, but the transition from successful research outcomes to production-scale deployment encounters two critical obstacles. First, the scarcity of labeled training data (i.e., identifying healthy and anomalous runs) in telemetry datasets makes it hard to train these ML systems effectively. Second, runtime analysis, required for providing timely detection and diagnosis of performance anomalies, demands seamless integration of ML-based methods with the monitoring frameworks. This thesis claims that ML-based performance analytics frameworks that leverage a limited amount of labeled data and ensure runtime analysis can achieve sufficient anomaly diagnosis performance for production HPC systems. To support this claim, we undertake ML-based performance analytics on two fronts. First, we design and develop novel frameworks for anomaly diagnosis that leverage semi-supervised or unsupervised learning techniques to reduce the need for extensive labeled data. Second, we design a simple yet adaptable architecture to enable deployment and demonstrate that these frameworks are feasible for runtime analysis. This thesis makes the following specific contributions: First, we design a semi-supervised anomaly diagnosis framework, Proctor, which operates with hundreds of labeled samples (in contrast to tens of thousands) and a vast number of unlabeled samples. We show that Proctor outperforms the fully supervised baseline by up to 11% in F1-score for diagnosing anomalies when there are approximately 30 labeled samples. We then reframe the problem and introduce ALBADRoss to determine which samples should be labeled by experts to maximize the model performance using active learning. On a production HPC dataset, ALBADRoss achieves a 0.95 F1-score (the same score that a fully-supervised framework achieved) and near-zero false alarm rate using 24x fewer labeled samples. Finally, with Prodigy, we solve the anomaly detection problem but with a focus on deployment. Prodigy is designed for detecting performance anomalies on compute nodes using unsupervised learning. Our framework achieves a 0.95 F1-score in detecting anomalies on a production HPC system telemetry dataset. We also design a simple and adaptable software architecture and deploy it on a 1488-node production HPC system, detecting real-world performance anomalies with 88% accuracy.

京都大学 / 新制・課程博士 / 博士(理学) / 甲第24411号 / 理博第4910号 / 新制||理||1702(附属図書館) / 京都大学大学院理学研究科物理学・宇宙物理学専攻 / (主査)教授 青木 慎也, 教授 大西 明, 教授 橋本 幸士 / 学位規則第4条第1項該当 / Doctor of Science / Kyoto University / DFAM

Due to the explosive growth in the size of scientific data sets, data-intensive computing is an emerging trend in computational science. HPC applications are generating and processing large amount of data ranging from terabytes (TB) to petabytes (PB). This new trend of growth in data for HPC applications has imposed challenges as to what is an appropriate parallel programming framework to efficiently process large data sets. In this work, we study the applicability of two programming models (MPI/MPI-IO and MapReduce) to a variety of I/O-intensive HPC applications ranging from simulations to analytics. We identify several performance and programmer productivity related limitations of these existing programming models, if used for I/O-intensive applications. We propose new frameworks which will improve both performance and programmer productivity for the emerging I/O-intensive applications. Message Passing Interface (MPI) is widely used for writing HPC applications. MPI/MPI- IO allows a fine-grained control of assigning data and task distribution. At the programming frameworks level, various optimizations have been proposed to improve the performance of MPI/MPI-IO function calls. These performance optimizations are provided as various function options to the programmers. In order to write an efficient code, they are required to know the exact usage of the optimization functions, hence programmer productivity is limited. We propose an abstraction called Reduced Function Set Abstraction (RFSA) for MPI-IO to reduce the number of I/O functions and provide methods to automate the selection of appropriate I/O function for writing HPC simulation applications. The purpose of RFSA is to hide the performance optimization functions from the application developer, and relieve the application developer from deciding on a specific function. The proposed set of functions relies on a selection algorithm to decide among the most common optimizations provided by MPI-IO. Additionally, many application scientists are looking to integrate data-intensive computing into computational-intensive High Performance Computing facilities, particularly for data analytics. We have observed several scientific applications which must migrate their data from an HPC storage system to a data-intensive one. There is a gap between the data semantics of HPC storage and data-intensive system, hence, once migrated, the data must be further refined and reorganized. This reorganization must be performed before existing data-intensive tools such as MapReduce can be effectively used to analyze data. This reorganization requires at least two complete scans through the data set and then at least one MapReduce program to prepare the data before analyzing it. Running multiple MapReduce phases causes significant overhead for the application, in the form of excessive I/O operations. For every MapReduce application that must be run in order to complete the desired data analysis, a distributed read and write operation on the file system must be performed. Our contribution is to extend Map-Reduce to eliminate the multiple scans and also reduce the number of pre-processing MapReduce programs. We have added additional expressiveness to the MapReduce language in our novel framework called MapReduce with Access Patterns (MRAP), which allows users to specify the logical semantics of their data such that 1) the data can be analyzed without running multiple data pre-processing MapReduce programs, and 2) the data can be simultaneously reorganized as it is migrated to the data-intensive file system. We also provide a scheduling mechanism to further improve the performance of these applications. The main contributions of this thesis are, 1) We implement a selection algorithm for I/O functions like read/write, merge a set of functions for data types and file views and optimize the atomicity function by automating the locking mechanism in RFSA. By running different parallel I/O benchmarks on both medium-scale clusters and NERSC supercomputers, we show an improved programmer productivity (35.7% on average). This approach incurs an overhead of 2-5% for one particular optimization, and shows performance improvement of 17% when a combination of different optimizations is required by an application. 2) We provide an augmented Map-Reduce system (MRAP), which consist of an API and corresponding optimizations i.e. data restructuring and scheduling. We have demonstrated up to 33% throughput improvement in one real application (read-mapping in bioinformatics), and up to 70% in an I/O kernel of another application (halo catalogs analytics). Our scheduling scheme shows performance improvement of 18% for an I/O kernel of another application (QCD analytics).

No description available.

No description available.

No description available.