Thermal-Aware Networked Many-Core Systems

Advancements in IC processing technology has led to the innovation and growth happening in the consumer electronics sector and the evolution of the IT infrastructure supporting this exponential growth. One of the most difficult obstacles to this growth is the removal of large amount of heatgenerated by the processing and communicating nodes on the system. The scaling down of technology and the increase in power density is posing a direct and consequential effect on the rise in temperature. This has resulted in the increase in cooling budgets, and affects both the life-time reliability and performance of the system. Hence, reducing on-chip temperatures has become a major design concern for modern microprocessors. This dissertation addresses the thermal challenges at different levels for both 2D planer and 3D stacked systems. It proposes a self-timed thermal monitoring strategy based on the liberal use of on-chip thermal sensors. This makes use of noise variation tolerant and leakage current based thermal sensing for monitoring purposes. In order to study thermal management issues from early design stages, accurate thermal modeling and analysis at design time is essential. In this regard, spatial temperature profile of the global Cu nanowire for on-chip interconnects has been analyzed. It presents a 3D thermal model of a multicore system in order to investigate the effects of hotspots and the placement of silicon die layers, on the thermal performance of a modern ip-chip package. For a 3D stacked system, the primary design goal is to maximise the performance within the given power and thermal envelopes. Hence, a thermally efficient routing strategy for 3D NoC-Bus hybrid architectures has been proposed to mitigate on-chip temperatures by herding most of the switching activity to the die which is closer to heat sink. Finally, an exploration of various thermal-aware placement approaches for both the 2D and 3D stacked systems has been presented. Various thermal models have been developed and thermal control metrics have been extracted. An efficient thermal-aware application mapping algorithm for a 2D NoC has been presented. It has been shown that the proposed mapping algorithm reduces the effective area reeling under high temperatures when compared to the state of the art.

Agent-based management systems for many-core platforms : rigorous design and efficient implementation

Due to various advantages such as flexibility, scalability and updatability, software intensive systems are increasingly embedded in everyday life. The constantly growing number of functions executed by these systems requires a high level of performance from the underlying platform. The main approach to incrementing performance has been the increase of operating frequency of a chip. However, this has led to the problem of power dissipation, which has shifted the focus of research to parallel and distributed computing. Parallel many-core platforms can provide the required level of computational power along with low power consumption. On the one hand, this enables parallel execution of highly intensive applications. With their computational power, these platforms are likely to be used in various application domains: from home use electronics (e.g., video processing) to complex critical control systems. On the other hand, the utilization of the resources has to be efficient in terms of performance and power consumption. However, the high level of on-chip integration results in the increase of the probability of various faults and creation of hotspots leading to thermal problems. Additionally, radiation, which is frequent in space but becomes an issue also at the ground level, can cause transient faults. This can eventually induce a faulty execution of applications. Therefore, it is crucial to develop methods that enable efficient as well as resilient execution of applications. The main objective of the thesis is to propose an approach to design agentbased systems for many-core platforms in a rigorous manner. When designing such a system, we explore and integrate various dynamic reconfiguration mechanisms into agents functionality. The use of these mechanisms enhances resilience of the underlying platform whilst maintaining performance at an acceptable level. The design of the system proceeds according to a formal refinement approach which allows us to ensure correct behaviour of the system with respect to postulated properties. To enable analysis of the proposed system in terms of area overhead as well as performance, we explore an approach, where the developed rigorous models are transformed into a high-level implementation language. Specifically, we investigate methods for deriving fault-free implementations from these models into, e.g., a hardware description language, namely VHDL.

Energy aware software for many-core systems

Many-core systems provide a great potential in application performance with the massively parallel structure. Such systems are currently being integrated into most parts of daily life from high-end server farms to desktop systems, laptops and mobile devices. Yet, these systems are facing increasing challenges such as high temperature causing physical damage, high electrical bills both for servers and individual users, unpleasant noise levels due to active cooling and unrealistic battery drainage in mobile devices; factors caused directly by poor energy efficiency. Power management has traditionally been an area of research providing hardware solutions or runtime power management in the operating system in form of frequency governors. Energy awareness in application software is currently non-existent. This means that applications are not involved in the power management decisions, nor does any interface between the applications and the runtime system to provide such facilities exist. Power management in the operating system is therefore performed purely based on indirect implications of software execution, usually referred to as the workload. It often results in over-allocation of resources, hence power waste. This thesis discusses power management strategies in many-core systems in the form of increasing application software awareness of energy efficiency. The presented approach allows meta-data descriptions in the applications and is manifested in two design recommendations: 1) Energy-aware mapping 2) Energy-aware execution which allow the applications to directly influence the power management decisions. The recommendations eliminate over-allocation of resources and increase the energy efficiency of the computing system. Both recommendations are fully supported in a provided interface in combination with a novel power management runtime system called Bricktop. The work presented in this thesis allows both new- and legacy software to execute with the most energy efficient mapping on a many-core CPU and with the most energy efficient performance level. A set of case study examples demonstrate realworld energy savings in a wide range of applications without performance degradation.

Gpu-based many-core architecture emulation: A double level approach

The efficient emulation of a many-core architecture is a challenging task, each core could be emulated through a dedicated thread and such threads would be interleaved on an either single-core or a multi-core processor. The high number of context switches will results in an unacceptable performance. To support this kind of application, the GPU computational power is exploited in order to schedule the emulation threads on the GPU cores. This presents a non trivial divergence issue, since GPU computational power is offered through SIMD processing elements, that are forced to synchronously execute the same instruction on different memory portions. Thus, a new emulation technique is introduced in order to overcome this limitation: instead of providing a routine for each ISA opcode, the emulator mimics the behavior of the Micro Architecture level, here instructions are date that a unique routine takes as input. Our new technique has been implemented and compared with the classic emulation approach, in order to investigate the chance of a hybrid solution.

Parallel Architectures for Many-Core Systems-On-Chip in Deep Sub-Micron Technology

Despite the several issues faced in the past, the evolutionary trend of silicon has kept its constant pace. Today an ever increasing number of cores is integrated onto the same die. Unfortunately, the extraordinary performance achievable by the many-core paradigm is limited by several factors. Memory bandwidth limitation, combined with inefficient synchronization mechanisms, can severely overcome the potential computation capabilities. Moreover, the huge HW/SW design space requires accurate and flexible tools to perform architectural explorations and validation of design choices. In this thesis we focus on the aforementioned aspects: a flexible and accurate Virtual Platform has been developed, targeting a reference many-core architecture. Such tool has been used to perform architectural explorations, focusing on instruction caching architecture and hybrid HW/SW synchronization mechanism. Beside architectural implications, another issue of embedded systems is considered: energy efficiency. Near Threshold Computing is a key research area in the Ultra-Low-Power domain, as it promises a tenfold improvement in energy efficiency compared to super-threshold operation and it mitigates thermal bottlenecks. The physical implications of modern deep sub-micron technology are severely limiting performance and reliability of modern designs. Reliability becomes a major obstacle when operating in NTC, especially memory operation becomes unreliable and can compromise system correctness. In the present work a novel hybrid memory architecture is devised to overcome reliability issues and at the same time improve energy efficiency by means of aggressive voltage scaling when allowed by workload requirements. Variability is another great drawback of near-threshold operation. The greatly increased sensitivity to threshold voltage variations in today a major concern for electronic devices. We introduce a variation-tolerant extension of the baseline many-core architecture. By means of micro-architectural knobs and a lightweight runtime control unit, the baseline architecture becomes dynamically tolerant to variations.

Many-Core Architectures: Hardware-Software Optimization and Modeling Techniques

During the last few decades an unprecedented technological growth has been at the center of the embedded systems design paramount, with Moore’s Law being the leading factor of this trend. Today in fact an ever increasing number of cores can be integrated on the same die, marking the transition from state-of-the-art multi-core chips to the new many-core design paradigm. Despite the extraordinarily high computing power, the complexity of many-core chips opens the door to several challenges. As a result of the increased silicon density of modern Systems-on-a-Chip (SoC), the design space exploration needed to find the best design has exploded and hardware designers are in fact facing the problem of a huge design space. Virtual Platforms have always been used to enable hardware-software co-design, but today they are facing with the huge complexity of both hardware and software systems. In this thesis two different research works on Virtual Platforms are presented: the first one is intended for the hardware developer, to easily allow complex cycle accurate simulations of many-core SoCs. The second work exploits the parallel computing power of off-the-shelf General Purpose Graphics Processing Units (GPGPUs), with the goal of an increased simulation speed. The term Virtualization can be used in the context of many-core systems not only to refer to the aforementioned hardware emulation tools (Virtual Platforms), but also for two other main purposes: 1) to help the programmer to achieve the maximum possible performance of an application, by hiding the complexity of the underlying hardware. 2) to efficiently exploit the high parallel hardware of many-core chips in environments with multiple active Virtual Machines. This thesis is focused on virtualization techniques with the goal to mitigate, and overtake when possible, some of the challenges introduced by the many-core design paradigm.

Many-core Algorithms for Combinatorial Optimization

Combinatorial Optimization is becoming ever more crucial, in these days. From natural sciences to economics, passing through urban centers administration and personnel management, methodologies and algorithms with a strong theoretical background and a consolidated real-word effectiveness is more and more requested, in order to find, quickly, good solutions to complex strategical problems. Resource optimization is, nowadays, a fundamental ground for building the basements of successful projects. From the theoretical point of view, Combinatorial Optimization rests on stable and strong foundations, that allow researchers to face ever more challenging problems. However, from the application point of view, it seems that the rate of theoretical developments cannot cope with that enjoyed by modern hardware technologies, especially with reference to the one of processors industry. In this work we propose new parallel algorithms, designed for exploiting the new parallel architectures available on the market. We found that, exposing the inherent parallelism of some resolution techniques (like Dynamic Programming), the computational benefits are remarkable, lowering the execution times by more than an order of magnitude, and allowing to address instances with dimensions not possible before. We approached four Combinatorial Optimization’s notable problems: Packing Problem, Vehicle Routing Problem, Single Source Shortest Path Problem and a Network Design problem. For each of these problems we propose a collection of effective parallel solution algorithms, either for solving the full problem (Guillotine Cuts and SSSPP) or for enhancing a fundamental part of the solution method (VRP and ND). We endorse our claim by presenting computational results for all problems, either on standard benchmarks from the literature or, when possible, on data from real-world applications, where speed-ups of one order of magnitude are usually attained, not uncommonly scaling up to 40 X factors.

Memory isolation in many-core embedded systems

The current approach to developing mixed-criticality sys- tems is by partitioning the hardware resources (processors, memory and I/O devices) among the different applications. Partitions are isolated from each other both in the temporal and the spatial domain, so that low-criticality applications cannot compromise other applications with a higher level of criticality in case of misbehaviour. New architectures based on many-core processors open the way to highly parallel systems in which each partition can be allocated to a set of dedicated proces- sor cores, thus simplifying partition scheduling and temporal separation. Moreover, spatial isolation can also benefit from many-core architectures, by using simpler hardware mechanisms to protect the address spaces of different applications. This paper describes an architecture for many- core embedded partitioned systems, together with some implementation advice for spatial isolation.

A Kinetic Vlasov Model for Plasma Simulation Using Discontinuous Galerkin Method on Many-Core Architectures

Thesis (Ph.D.)--University of Washington, 2016-03

Optimization of Pattern Matching Algorithms for Multi- and Many-Core Platforms

Image and video compression play a major role in the world today, allowing the storage and transmission of large multimedia content volumes. However, the processing of this information requires high computational resources, hence the improvement of the computational performance of these compression algorithms is very important. The Multidimensional Multiscale Parser (MMP) is a pattern-matching-based compression algorithm for multimedia contents, namely images, achieving high compression ratios, maintaining good image quality, Rodrigues et al. [2008]. However, in comparison with other existing algorithms, this algorithm takes some time to execute. Therefore, two parallel implementations for GPUs were proposed by Ribeiro [2016] and Silva [2015] in CUDA and OpenCL-GPU, respectively. In this dissertation, to complement the referred work, we propose two parallel versions that run the MMP algorithm in CPU: one resorting to OpenMP and another that converts the existing OpenCL-GPU into OpenCL-CPU. The proposed solutions are able to improve the computational performance of MMP by 3 and 2:7 , respectively. The High Efficiency Video Coding (HEVC/H.265) is the most recent standard for compression of image and video. Its impressive compression performance, makes it a target for many adaptations, particularly for holoscopic image/video processing (or light field). Some of the proposed modifications to encode this new multimedia content are based on geometry-based disparity compensations (SS), developed by Conti et al. [2014], and a Geometric Transformations (GT) module, proposed by Monteiro et al. [2015]. These compression algorithms for holoscopic images based on HEVC present an implementation of specific search for similar micro-images that is more efficient than the one performed by HEVC, but its implementation is considerably slower than HEVC. In order to enable better execution times, we choose to use the OpenCL API as the GPU enabling language in order to increase the module performance. With its most costly setting, we are able to reduce the GT module execution time from 6.9 days to less then 4 hours, effectively attaining a speedup of 45 .

Application mapping in NoC-Based many-cores

Many-core platforms based on Network-on-Chip (NoC [Benini and De Micheli 2002]) present an emerging technology in the real-time embedded domain. Although the idea to group the applications previously executed on separated single-core devices, and accommodate them on an individual many-core chip offers various options for power savings, cost reductions and contributes to the overall system flexibility, its implementation is a non-trivial task. In this paper we address the issue of application mapping onto a NoCbased many-core platform when considering fundamentals and trends of current many-core operating systems, specifically, we elaborate on a limited migrative application model encompassing a message-passing paradigm as a communication primitive. As the main contribution, we formulate the problem of real-time application mapping, and propose a three-stage process to efficiently solve it. Through analysis it is assured that derived solutions guarantee the fulfilment of posed time constraints regarding worst-case communication latencies, and at the same time provide an environment to perform load balancing for e.g. thermal, energy, fault tolerance or performance reasons.We also propose several constraints regarding the topological structure of the application mapping, as well as the inter- and intra-application communication patterns, which efficiently solve the issues of pessimism and/or intractability when performing the analysis.

Real-Time Application Mapping for Many-Cores Using a Limited Migrative Model

Many-core platforms are an emerging technology in the real-time embedded domain. These devices offer various options for power savings, cost reductions and contribute to the overall system flexibility, however, issues such as unpredictability, scalability and analysis pessimism are serious challenges to their integration into the aforementioned area. The focus of this work is on many-core platforms using a limited migrative model (LMM). LMM is an approach based on the fundamental concepts of the multi-kernel paradigm, which is a promising step towards scalable and predictable many-cores. In this work, we formulate the problem of real-time application mapping on a many-core platform using LMM, and propose a three-stage method to solve it. An extended version of the existing analysis is used to assure that derived mappings (i) guarantee the fulfilment of timing constraints posed on worst-case communication delays of individual applications, and (ii) provide an environment to perform load balancing for e.g. energy/thermal management, fault tolerance and/or performance reasons.

Use of shared memory in the context of embedded multi-core processors: exploration of the technology and its limits

Modern embedded systems embrace many-core shared-memory designs. Due to constrained power and area budgets, most of them feature software-managed scratchpad memories instead of data caches to increase the data locality. It is therefore programmers’ responsibility to explicitly manage the memory transfers, and this make programming these platform cumbersome. Moreover, complex modern applications must be adequately parallelized before they can the parallel potential of the platform into actual performance. To support this, programming languages were proposed, which work at a high level of abstraction, and rely on a runtime whose cost hinders performance, especially in embedded systems, where resources and power budget are constrained. This dissertation explores the applicability of the shared-memory paradigm on modern many-core systems, focusing on the ease-of-programming. It focuses on OpenMP, the de-facto standard for shared memory programming. In a first part, the cost of algorithms for synchronization and data partitioning are analyzed, and they are adapted to modern embedded many-cores. Then, the original design of an OpenMP runtime library is presented, which supports complex forms of parallelism such as multi-level and irregular parallelism. In the second part of the thesis, the focus is on heterogeneous systems, where hardware accelerators are coupled to (many-)cores to implement key functional kernels with orders-of-magnitude of speedup and energy efficiency compared to the “pure software” version. However, three main issues rise, namely i) platform design complexity, ii) architectural scalability and iii) programmability. To tackle them, a template for a generic hardware processing unit (HWPU) is proposed, which share the memory banks with cores, and the template for a scalable architecture is shown, which integrates them through the shared-memory system. Then, a full software stack and toolchain are developed to support platform design and to let programmers exploiting the accelerators of the platform. The OpenMP frontend is extended to interact with it.

Towards network-on-chip agreement protocols

Demands for functionality enhancements, cost reductions and power savings clearly suggest the introduction of multiand many-core platforms in real-time embedded systems. However, when compared to uni-core platforms, the manycores experience additional problems, namely the lack of scalable coherence mechanisms and the necessity to perform migrations. These problems have to be addressed before such systems can be considered for integration into the realtime embedded domain. We have devised several agreement protocols which solve some of the aforementioned issues. The protocols allow the applications to plan and organise their future executions both temporally and spatially (i.e. when and where the next job will be executed). Decisions can be driven by several factors, e.g. load balancing, energy savings and thermal issues. All presented protocols are analytically described, with the particular emphasis on their respective real-time behaviours and worst-case performance. The underlying assumptions are based on the multi-kernel model and the message-passing paradigm, which constitutes the communication between the interacting instances.