No description available.

No description available.

Understanding user and system behavior is most vital for designing efficient systems. Most systems are designed with certain user workload in mind. However, such workloads evolve over time, or the underlying hardware assumptions change. Further, most modern systems are not built or deployed in isolation, they interact with other systems whose behavior might not be exactly understood. Thus in order to understand the performance of a system, it must be inspected closely while user workloads are running. Such close inspection must be done with minimum disturbance to the user workload. Thus tracing or collection of all the user and system generated events becomes an important approach in gaining comprehensive insight in user behavior. As part of this work, we have three major contributions. We designed and implemented an in-depth block level I/O tracer, which would collect block level information like sector number, size of the I/O, actual contents of the I/O, along with certain file system information like filename, and offset in the file, for every I/O request. Next, to minimize the impact of the tracing to the running workload, we introduce and implement a sampling mechanism which traces fewer I/O requests. We validate that this sampling preserves certain I/O access patterns. Finally, as one of the application of our tracer, we use it as a crucial component of a system designed to do VM placements according to user workload. / Master of Science

The demands of exascale computing systems and applications have pushed for a rapid, continual design paradigm coupled with increasing design complexities from the interaction between the application, the middleware, and the underlying system hardware, which forms a breeding ground for inefficiency. This work seeks to improve system efficiency by exposing the root causes of unexpected performance slowdowns (e.g., lower performance at higher processor speeds) that occur more frequently in power-scalable systems where raw processor speed varies. More precisely, we perform an exhaustive empirical study that conclusively shows that increasing processor speed often reduces performance and wastes energy. Our experimental work shows that the frequency of occurrence and magnitude of slowdowns grow with clock frequency and parallelism, indicating that such slowdowns will increasingly be observed with trends in processor and system design. Performance speedups at lower frequencies (or slowdowns at higher frequencies) have been anecdotally observed in the prevailing literature since 2004, but no research has explained nor exploited this phenomenon. This work conclusively demonstrates that performance slowdowns during processor speedup phases can exceed 47% in common I/O workloads. Our hypothesis challenges (and ultimately debunks) a fundamental assumption in computer systems: faster processor speeds result in the same or better performance. In this work, with the use of code and kernel instrumentation, exhaustive experiments, and deep insight into the inner workings of the Linux I/O subsystem, I overcome the aforementioned challenges of variance, complexity, and nondeterminism and identify the I/O resource contention as the root cause of the slowdowns during processor speedup. Specifically, such contention comes from the Linux kernel when the journaling block device (JBD) interacts with the ext3/4 file system that introduces file write delays and file synchronization delays. To fully explain how such I/O contention causes performance anomaly, I propose analytical models of resource contention among I/O threads to describe the root cause of the observed I/O slowdowns when processors speed up. To this end, I introduce LUC, a runtime system to limit the unintended consequences of power scaling and demonstrate the effectiveness of the LUC system for two critical parallel transaction-oriented workloads, including a mail server (varMail) and online transaction processing (oltp). / Ph. D.

Uppsatsen beskriver arbetet och undersökning för hur deduplicering i filsystemet ZFS påverkar effektförbrukningen. En större mängd redundant data förekommer i centraliserade lagringssystem som förser virtualiserade servrar med lagringsutrymme. Deduplicering kan för den typen av lagringsmiljö eliminera redundant data och ger en stor besparing av lagringsutrymme. Frågan som undersökningen avsåg att besvara var hur ett lagringssystem påverkas av det extra arbete som det innebär att deduplicera data i realtid.Metoden för att undersöka problemet var att utföra fem experiment med olika typer av scenarion. Varje scenario innebar att filer kopierades till ett lagringssystem med eller utan deduplicering för att senare kunna analysera skillnaden. Dessutom varierades mängden deduplicerbar data under experimenten vilket skulle visa om belastningen på hårddiskarna förändrades.Resultatet av experimenten visar att deduplicering ökar effektförbrukning och processorbelastning medan antalet I/O-operationer minskar. Analysen av resultatet visar att med en stigande andel deduplicerbar data som skrivs till hårddiskarna så stiger också effektförbrukning och processorbelastning. / This report describes the process and outcome of the research on how the power consumption is affected by deduplication in a ZFS file system. A large amount of redundant data exists in centralized storage systems that provide virtualized servers with storage space. Deduplication can be used to eliminate redundant data and give an improved utilization of available space in this kind of storage environment. The question that the study sought to answer was how a storage systems power consumption is affected by the extra workload deduplication introduces.The method used to investigate the problem was to perform five experiments with different types of scenarios. The difference in each scenario was that the data was written to a storage system with or without deduplication to later analyze the difference. Each scenario had a varied amount deduplicatable data during the experiments which would show if the load on disks changed.The results show that deduplication increases the power consumption and CPU load while the I/O-operations decrease. The analysis of the result shows that increasing the deduplicatable data also increases the power consumption and CPU load.

This work describes development of control hardware and software for automatic production of tuning fork with tip for atomic force microscopy (AFM). A specialized device implementing the problem is developed and tested. The advantage of the proposed solution is the ability to produce the tuning forks with constant parameters, which is an important condition for successful work with the microscope.

Vorgestellt wird die Architektur des Data Centers der TU Chemnitz, die auf den neuen Technologie Data Center Bridging (DCB) und Fibre Channel over Ethernet (FCoE) basiert. Es werden die entsprechenden Standards dargestellt und ein Überblick zur gegenwärtig verfügbaren Technik gegeben. Das Rechenzentrum der TU Chemnitz setzt diese Technologien bereits erfolgreich im Kontext von VMware-Virtualisierung und bei Betrieb I/O-intensiver Systeme ein.

To transfer data between devices and main memory, standard C block I/O interfaces use block buffers of type char. C++ programs that perform block I/O commonly use typecasting to move data between structures and block buffers. The subject of this thesis, the layout description language (LADEL), represents a high-level solution to the problem of block buffer management. LADEL provides operators that hide the casting ordinarily required to pack and to unpack buffers and guard against overflow of the virtual fields. LADEL also allows a programmer to dynamically define a structured view of a block buffer's contents. This view includes the use of variable length field specifiers, which supports the development of a general specification for an I/O block that optimizes the use of preset buffers. The need for optimizing buffer use arises in file processing algorithms that perform optimally when I/O buffers are filled to capacity. Packing a buffer to capacity can require reasonably complex C++ code. LADEL can be used to reduce this complexity to considerable extent. C++ programs written using LADEL are less complex, easy to maintain, and easier to read than equivalent programs written LADEL. This increase in maintainability is achieved at a cost of approximately 11 % additional time in comparison to programs that use casting to manipulate block buffer data.

No description available.

The increasing size of high performance computing systems and the associated increase in the volume of generated data, has resulted in an I/O bottleneck for these applications. This bottleneck is further exacerbated by the imbalance in the growth of processing capability compared to storage capability, due mainly to the power and cost requirements of scaling the storage. This thesis introduces data services, a new abstraction which provides significant benefits for data intensive applications. Data services combine low overhead data movement with flexible placement of data manipulation operations, to address the I/O challenges of leadership class scientific applications. The impact of asynchronous data movement on application runtime is minimized by utilizing novel server side data movement schedulers to avoid contention related jitter in application communication. Additionally, the JITStager component is presented. Utilizing dynamic code generation and flexible code placement, the JITStager allows data services to be executed as a pipeline extending from the application to storage. It is shown in this thesis that data services can add new functionality to the application without having an significant negative impact on performance.