The advances in computing and communication technologies and software tools have resulted in an explosive growth in networked applications and information services that cover all aspects of our life. These services and applications are inherently complex, dynamic and heterogeneous. In a similar way, the underlying information infrastructure, e.g. the Internet, is large, complex, heterogeneous and dynamic, globally aggregating large numbers of independent computing and communication resources. The combination of the two results in application development and management complexities that break current computing paradigms, which are based on static behaviors. As a result, applications, programming environments and information infrastructures are rapidly becoming fragile, unmanageable and insecure. This has led researchers to consider alternative programming paradigms and management techniques that are based on strategies used by biological systems. Autonomic programming paradigm is inspired by the human autonomic nervous system that handles complexity, uncertainties and abnormality. The overarching goal of the autonomic programming paradigm is to help building systems and applications capable of self-management. Firstly, we investigated the large-scale scientific computing applications which generally experience different execution phases at run time and each phase has different computational, communication and storage requirements as well as different physical characteristics. In this dissertation, we present Physics Aware Optimization (PAO) paradigm that enables programmers to identify the appropriate solution methods to exploit the heterogeneity and the dynamism of the application execution states. We implement a Physics Aware Optimization Manager to exploit the PAO paradigm. On the other hand we present a self configuration paradigm based on the principles of autonomic computing that can handle efficiently complexity, dynamism and uncertainty in configuring server and networked systems and their applications. Our approach is based on making any resource/application to operate as an Autonomic Component (that means it can be self-managed component) by using our autonomic programming paradigm. Our POA technique for medical application yielded about 3X improvement of performance with 98.3% simulation accuracy compared to traditional techniques for performance optimization. Also, our Self-configuration management for power and performance management in GPU cluster demonstrated 53.7% power savings for CUDAworkload while maintaining the cluster performance within given acceptable thresholds.

High-Performance Computing (HPC) resources, such as Supercomputers, Clusters, Grids and HPC Clouds, are managed by Resource Management Systems (RMSs) that multiple resources among multiple users and decide how computing nodes are allocated to user applications. As more and more petascale computing resources are built and exascale is to be achieved by 2020, optimizing resource allocation to applications is critical to ensure their efficient execution. However, current RMSs, such as batch schedulers, only offer a limited interface. In most cases, the application has to blindly choose resources at submittal without being able to adapt its choice to the state of the target resources, neither before it started nor during execution. The goal of this Thesis is to improve resource management, so as to allow applications to efficiently allocate resources. We achieve this by proposing software architectures that promote collaboration between the applications and the RMS, thus, allowing applications to negotiate the resources they run on. To this end, we start by analysing the various types of applications and their unique resource requirements, categorizing them into rigid, moldable, malleable and evolving. For each case, we highlight the opportunities they open up for improving resource management.The first contribution deals with moldable applications, for which resources are only negotiated before they start. We propose CooRMv1, a centralized RMS architecture, which delegates resource selection to the application launchers. Simulations show that the solution is both scalable and fair. The results are validated through a prototype implementation deployed on Grid'5000. Second, we focus on negotiating allocations on geographically-distributed resources, managed by multiple institutions. We build upon CooRMv1 and propose distCooRM, a distributed RMS architecture, which allows moldable applications to efficiently co-allocate resources managed by multiple independent agents. Simulation results show that distCooRM is well-behaved and scales well for a reasonable number of applications. Next, attention is shifted to run-time negotiation of resources, so as to improve support for malleable and evolving applications. We propose CooRMv2, a centralized RMS architecture, that enables efficient scheduling of evolving applications, especially non-predictable ones. It allows applications to inform the RMS about their maximum expected resource usage, through pre-allocations. Resources which are pre-allocated but unused can be filled by malleable applications. Simulation results show that considerable gains can be achieved. Last, production-ready software are used as a starting point, to illustrate the interest as well as the difficulty of improving cooperation between existing systems. GridTLSE is used as an application and DIET as an RMS to study a previously unsupported use-case. We identify the underlying problem of scheduling optional computations and propose an architecture to solve it. Real-life experiments done on the Grid'5000 platform show that several metrics are improved, such as user satisfaction, fairness and the number of completed requests. Moreover, it is shown that the solution is scalable.

The CellPilot library provides a uniform communication programming model, based on Pilot's process/channel approach, for clusters of Cell Broadband Engine processors. Pilot, a thin layer on top of the Message Passing Interface (MPI) library, allows processes to read/write messages on channels defined between pairs of processes on the cluster, but Pilot alone does not help a Cell programmer cope with the considerable complexities of intra-Cell communication. With CellPilot, programmers still design software in terms of processes, but they can now be located on a Cell node's Power Processor Elements (PPEs), Synergistic Processing Elements (SPEs), or non-Cell node within a heterogeneous Cell cluster, and communication is accomplished via channels between process pairs. Programs are coded in terms of reading and writing on those channels, whereupon CellPilot transparently applies whichever communication mechanisms are required to transport the message, regardless of its endpoints. This gives the programmer a way to handle inter-process communication while avoiding low-level I/O operations and the use of multiple libraries.

Big Data can be characterized by 3 V's. * Big Volume refers to the unprecedented growth in the amount of data. * Big Velocity refers to the growth in the speed of moving data in and out management systems. * Big Variety refers to the growth in the number of different data formats. Managing Big Data requires fundamental changes in the architecture of data management systems. Data storage should continue being innovated in order to adapt to the growth of data. They need to be scalable while maintaining high performance regarding data accesses. This thesis focuses on building scalable data management systems for Big Data. Our first and second contributions address the challenge of providing efficient support for Big Volume of data in data-intensive high performance computing (HPC) environments. Particularly, we address the shortcoming of existing approaches to handle atomic, non-contiguous I/O operations in a scalable fashion. We propose and implement a versioning-based mechanism that can be leveraged to offer isolation for non-contiguous I/O without the need to perform expensive synchronizations. In the context of parallel array processing in HPC, we introduce Pyramid, a large-scale, array-oriented storage system. It revisits the physical organization of data in distributed storage systems for scalable performance. Pyramid favors multidimensional-aware data chunking, that closely matches the access patterns generated by applications. Pyramid also favors a distributed metadata management and a versioning concurrency control to eliminate synchronizations in concurrency. Our third contribution addresses Big Volume at the scale of the geographically distributed environments. We consider BlobSeer, a distributed versioning-oriented data management service, and we propose BlobSeer-WAN, an extension of BlobSeer optimized for such geographically distributed environments. BlobSeer-WAN takes into account the latency hierarchy by favoring locally metadata accesses. BlobSeer-WAN features asynchronous metadata replication and a vector-clock implementation for collision resolution. To cope with the Big Velocity characteristic of Big Data, our last contribution feautures DStore, an in-memory document-oriented store that scale vertically by leveraging large memory capability in multicore machines. DStore demonstrates fast and atomic complex transaction processing in data writing, while maintaining high throughput read access. DStore follows a single-threaded execution model to execute update transactions sequentially, while relying on a versioning concurrency control to enable a large number of simultaneous readers.

<p>Self-tuning has been an elusive goal for operating systems and is becoming a pressing issue for modern operating systems. Well-trained system administrators are able to tune an operating system to achieve better system performance for a specific system class. Unfortunately, the system class can change when the running applications change. The model for self-tuning operating system is based on a monitor-classify-adjust loop. The idea of this loop is to continuously monitor certain performance metrics, and whenever these change, the system determines the new system class and dynamically adjusts tuning parameters for this new class. This thesis described KernTune, a prototype tool that identifies the system class and improves system performance automatically. A key aspect of KernTune is the notion of Artificial Intelligence oriented performance tuning. Its uses a support vector machine to identify the system class, and tunes the operating system for that specific system class. This thesis presented design and implementation details for KernTune. It showed how KernTune identifies a system class and tunes the operating system for improved performance.</p>

No description available.

Parallel computing can take many forms. From a user's perspective, it is important to consider the advantages and disadvantages of each methodology. The following project attempts to provide some perspective on the methods of parallel computing and indicate where the tradeoffs lie along the continuum. Problems that are parallelizable enable researchers to maximize the computing resources available for a problem, and thus push the limits of the problems that can be solved. Solving any particular problem in parallel will require some very important design decisions to be made. These decisions may dramatically affect portability, performance, and cost of implementing a software solution to the problem. The results gained from this work indicate that although performance improvements are indeed possible---they are heavily dependent on the application in question and may require much more programming effort and expertise to implement.

This work endeavours to achieve technology transfer between established best-practice in academic high-performance computing and current techniques in commercial high-productivity computing. It shows that a credible high-performance message-passing communication library, with semantics and syntax following the Message-Passing Interface (MPI) Standard, can be built in pure C# (one of the .Net suite of computer languages). Message-passing has been the dominant paradigm in high-performance parallel programming of distributed-memory computer architectures for three decades. The MPI Standard originally distilled architecture-independent and language-agnostic ideas from existing specialised communication libraries and has since been enhanced and extended. Object-oriented languages can increase programmer productivity, for example by allowing complexity to be managed through encapsulation. Both the C# computer language and the .Net common language runtime (CLR) were originally developed by Microsoft Corporation but have since been standardised by the European Computer Manufacturers Association (ECMA) and the International Standards Organisation (ISO), which facilitates portability of source-code and compiled binary programs to a variety of operating systems and hardware. Combining these two open and mature technologies enables mainstream programmers to write tightly-coupled parallel programs in a popular standardised object-oriented language that is portable to most modern operating systems and hardware architectures. This work also establishes that a thread-to-thread delivery option increases shared-memory communication performance between MPI ranks on the same node. This suggests that the thread-as-rank threading model should be explicitly specified in future versions of the MPI Standard and then added to existing MPI libraries for use by thread-safe parallel codes. This work also ascertains that the C# socket object suffers from undesirable characteristics that are critical to communication performance and proposes ways of improving the implementation of this object.

Scientific simulations running on High End Computing machines in domains like Fusion, Astrophysics, and Combustion now routinely generate terabytes of data in a single run, and these data volumes are only expected to increase. Since such massive simulation outputs are key to scientific discovery, the ability to rapidly store, move, analyze, and visualize data is critical to scientists' productivity. Yet there are already serious I/O bottlenecks on current supercomputers, and movement toward the Exascale is further accelerating this trend. This dissertation is concerned with the design, implementation, and evaluation of middleware-level solutions to enable high performance and resource efficient online data analytics to process massive simulation output data at large scales. Online data analytics can effectively overcome the I/O bottleneck for scientific applications at large scales by processing data as it moves through the I/O path. Online analytics can extract valuable insights from live simulation output in a timely manner, better prepare data for subsequent deep analysis and visualization, and gain improved performance and reduced data movement cost (both in time and in power) compared to the conventional post-processing paradigm. The thesis identifies the key challenges for online data analytics based on the needs of a variety of large-scale scientific applications, and proposes a set of novel and effective approaches to efficiently program, distribute, and schedule online data analytics along the critical I/O path. In particular, its solution approach i) provides a high performance data movement substrate to support parallel and complex data exchanges between simulation and online data analytics, ii) enables placement flexibility of analytics to exploit distributed resources, iii) for co-placement of analytics with simulation codes on the same nodes, it uses fined-grained scheduling to harvest idle resources for running online analytics with minimal interference to the simulation, and finally, iv) it supports scalable efficient online spatial indices to accelerate data analytics and visualization on the deep memory hierarchies of high end machines. Our middleware approach is evaluated with leadership scientific applications in domains like Fusion, Combustion, and Molecular Dynamics, and on different High End Computing platforms. Substantial improvements are demonstrated in end-to-end application performance and in resource efficiency at scales of up to 16384 cores, for a broad range of analytics and visualization codes. The outcome is a useful and effective software platform for online scientific data analytics facilitating large-scale scientific data exploration.

This thesis presents a top to bottom analysis on designing and implementing fast algorithms for current and future systems. We present new analysis, algorithmic techniques, and implementations of the Fast Multipole Method (FMM) for solving N- body problems. We target the FMM because it is broadly applicable to a variety of scientific particle simulations used to study electromagnetic, fluid, and gravitational phenomena, among others. Importantly, the FMM has asymptotically optimal time complexity with guaranteed approximation accuracy. As such, it is among the most attractive solutions for scalable particle simulation on future extreme scale systems. We specifically address two key challenges. The first challenge is how to engineer fast code for today’s platforms. We present the first in-depth study of multicore op- timizations and tuning for FMM, along with a systematic approach for transforming a conventionally-parallelized FMM into a highly-tuned one. We introduce novel opti- mizations that significantly improve the within-node scalability of the FMM, thereby enabling high-performance in the face of multicore and manycore systems. The second challenge is how to understand scalability on future systems. We present a new algorithmic complexity analysis of the FMM that considers both intra- and inter- node communication costs. Using these models, we present results for choosing the optimal algorithmic tuning parameter. This analysis also yields the surprising prediction that although the FMM is largely compute-bound today, and therefore highly scalable on current systems, the trajectory of processor architecture designs, if there are no significant changes could cause it to become communication-bound as early as the year 2015. This prediction suggests the utility of our analysis approach, which directly relates algorithmic and architectural characteristics, for enabling a new kind of highlevel algorithm-architecture co-design. To demonstrate the scientific significance of FMM, we present two applications namely, direct simulation of blood which is a multi-scale multi-physics problem and large-scale biomolecular electrostatics. MoBo (Moving Boundaries) is the infrastruc- ture for the direct numerical simulation of blood. It comprises of two key algorithmic components of which FMM is one. We were able to simulate blood flow using Stoke- sian dynamics on 200,000 cores of Jaguar, a peta-flop system and achieve a sustained performance of 0.7 Petaflop/s. The second application we propose as future work in this thesis is biomolecular electrostatics where we solve for the electrical potential using the boundary-integral formulation discretized with boundary element methods (BEM). The computational kernel in solving the large linear system is dense matrix vector multiply which we propose can be calculated using our scalable FMM. We propose to begin with the two dielectric problem where the electrostatic field is cal- culated using two continuum dielectric medium, the solvent and the molecule. This is only a first step to solving biologically challenging problems which have more than two dielectric medium, ion-exclusion layers, and solvent filled cavities. Finally, given the difficulty in producing high-performance scalable code, productivity is a key concern. Recently, numerical algorithms are being redesigned to take advantage of the architectural features of emerging multicore processors. These new classes of algorithms express fine-grained asynchronous parallelism and hence reduce the cost of synchronization. We performed the first extensive performance study of a recently proposed parallel programming model, called Concurrent Collections (CnC). In CnC, the programmer expresses her computation in terms of application-specific operations, partially-ordered by semantic scheduling constraints. The CnC model is well-suited to expressing asynchronous-parallel algorithms, so we evaluate CnC using two dense linear algebra algorithms in this style for execution on state-of-the-art mul- ticore systems. Our implementations in CnC was able to match and in some cases even exceed competing vendor-tuned and domain specific library codes. We combine these two distinct research efforts by expressing FMM in CnC, our approach tries to marry performance with productivity that will be critical on future systems. Looking forward, we would like to extend this to distributed memory machines, specifically implement FMM in the new distributed CnC, distCnC to express fine-grained paral- lelism which would require significant effort in alternative models.