Relentless CMOS scaling coupled with lower design tolerances is making ICs increasingly susceptible to transient faults, wear-out related permanent faults and process variations. Decreasing CMOS reliability implies that high-availability systems which were previously restricted to the domain of mainframe computers or specially designed fault-tolerant systems may be come important for the commodity market as well. In this thesis we tackle the problem of enabling efficient, low cost and configurable fault-tolerance using Chip Multiprocessors (CMPs). Our work studies architectural fault detection methods based on redundant execution, specifically focusing on “leader-follower” architectures. In such architectures redundant execution is performed on two cores/threads of a CMP. One thread acts as the leading thread while the other acts as the trailing thread. The leading thread assists the execution of the trailing thread by forwarding the results of its execution. These forwarded results are used as predictions in the trailing thread and help improve its performance. In this thesis, we introduce a new form of execution assistance called critical value forwarding. Critical value forwarding uses heuristics to identify instructions on the critical path of execution and forwards the results of these instructions to the trailing core. The advantage of critical value forwarding is that it provides much of the speed up obtained by forwarding all values at a fraction of the bandwidth cost. We propose two architectures to exploit the idea of critical value forwarding. The first of these operates the trailing core at lower voltage/frequency levels in order to provide energy-efficient redundant execution. In this context, we also introduce algorithms to dynamically adapt the voltage/frequency level of the trailing core based on program behavior. Our experimental evaluation shows that this proposal consumes only 1.26 times the energy of a non-fault-tolerant baseline and has a mean performance overhead of about 1%. We compare our proposal to two previous energy-efficient fault-tolerant CMP proposals and find that our proposal delivers higher energy-efficiency and lower performance degradation than both while providing a similar level of fault coverage. Our second proposal uses the idea of critical value forwarding to improve fault-tolerant CMP throughput. This is done by using coarse-grained multithreading to mul-tiplex trailing threads on a single core. Our evaluation shows that this architecture delivers 9–13% higher throughput than previous proposals, including one configuration that uses simultaneous multithreading(SMT) to multiplex trailing threads. Since this proposal increases fault-tolerant CMP throughput by executing multiple threads on a single core, it comes at a modest cost in single-threaded performance, a mean slowdown between11–14%.

La nueva tendencia para aumentar el rendimiento de los futuroscomputadores son los multiprocesadores en un solo chip (CMPs). Seespera que en un futuro cercano salgan al mercado CMPs con decenas deprocesadores. Hoy en d�a, la mejor manera de mantener la coherencia decache en estos sistemas es mediante los protocolos basados endirectorio. Sin embargo, estos protocolos tienen dos grandesproblemas: una gran sobrecarga de memoria y una alta latencia de losfallos de cache.Esta tesis se ha centrado en estos problemas claves para la eficienciay escalabilidad del CMP. En primer lugar, se ha presentado unaorganizaci�n de directorios escalable. En segundo lugar, se hanpropuesto los protocolos de coherencia directa, que evitan laindirecci�n al nodo home y, por tanto, reducen el tiempo de ejecuci�nde las aplicaciones. Por �ltimo, se ha desarrollado una pol�tica demapeo para caches compartidas pero f�sicamente distribuidas, quereduce la latencia de acceso y garantiza una distribuci�n uniforme delos datos con el fin de reducir su tasa de fallos. Esto se traducefinalmente en un menor tiempo de ejecuci�n para las aplicaciones. / Chip multiprocessors (CMPs) constitute the new trend for increasingthe performance of future computers. In the near future, chips withtens of cores will become more popular. Nowadays, directory-basedprotocols constitute the best alternative to keep cache coherence inlarge-scale systems. Nevertheless, directory-based protocols have twoimportant issues that prevent them from achieving better scalability:the directory memory overhead and the long cache miss latencies.This thesis focuses on these key issues. The first proposal is ascalable distributed directory organization that copes with the memoryoverhead of directory-based protocols. The second proposal presentsthe direct coherence protocols, which are aimed at avoiding theindirection problem of traditional directory-based protocols and,therefore, they improve applications' performance. Finally, a novelmapping policy for distributed caches is presented. This policyreduces the long access latency while lessening the number of off-chipaccesses, leading to improvements in applications' execution time.

Scientific computing is used frequently in an increasing number of disciplines to accelerate scientific discovery. Many such computing problems involve the numerical solution of partial differential equations (PDE). In this thesis we explore and develop methodology for high-performance implementations of PDE solvers for shared-memory multiprocessor architectures. We consider three realistic PDE settings: solution of the Maxwell equations in 3D using an unstructured grid and the method of conjugate gradients, solution of the Poisson equation in 3D using a geometric multigrid method, and solution of an advection equation in 2D using structured adaptive mesh refinement. We apply software optimization techniques to increase both parallel efficiency and the degree of data locality. In our evaluation we use several different shared-memory architectures ranging from symmetric multiprocessors and distributed shared-memory architectures to chip-multiprocessors. For distributed shared-memory systems we explore methods of data distribution to increase the amount of geographical locality. We evaluate automatic and transparent page migration based on runtime sampling, user-initiated page migration using a directive with an affinity-on-next-touch semantic, and algorithmic optimizations for page-placement policies. Our results show that page migration increases the amount of geographical locality and that the parallel overhead related to page migration can be amortized over the iterations needed to reach convergence. This is especially true for the affinity-on-next-touch methodology whereby page migration can be initiated at an early stage in the algorithms. We also develop and explore methodology for other forms of data locality and conclude that the effect on performance is significant and that this effect will increase for future shared-memory architectures. Our overall conclusion is that, if the involved locality issues are addressed, the shared-memory programming model provides an efficient and productive environment for solving many important PDE problems.

The number of processor cores and on-chip cache size has been increasing on chip multiprocessors (CMPs). As a result, leakage power dissipated in the on-chip cache has become very significant. We explore various techniques to switch-off the over-allocated cache so as to reduce leakage power consumed by it. A large cache offers non-uniform access latency to different cores present on a CMP and such a cache is called “Non-Uniform Cache Architecture (NUCA)”. Past studies have explored techniques to reduce leakage power for uniform access latency caches and with a single application executing on a uniprocessor. Our ideas of power optimized caches are applicable to any memory technology and architecture for which the difference of leakage power in the on-state and off-state of on-chip cache bank is significant. Switching off the last level shared cache on a CMP is a challenging problem due to concurrently executing threads/processes and large dispersed NUCA cache. Hence, to determine cache requirement on a CMP, first we propose a new highly accurate method to estimate working set size of an application, which we call “tagged working set size estimation (TWSS)” method. This method has a negligible hardware storage overhead of 0.1% of the cache size. The use of TWSS is demonstrated by adaptively adjusting cache associativity. Our ideas of adaptable associative cache is scalable with respect to the number of cores present on a CMP. It uses information available locally in a tile on a tiled CMP and thus avoids network access unlike other commonly used heuristics such as average memory access latency and cache miss ratio. Our implementation gives 25% and 19% higher EDP savings than that obtained with average memory access latency and cache miss ratio heuristics on a static NUCA platform (SNUCA), respectively. Cache misses increase with reduced cache associativity. Hence, we also propose to map some of the L2 slices onto the rest L2 slices and switch-off mapped L2 slices. The L2 slice includes all L2 banks in a tile. We call this technique the “remap policy”. Some applications execute with lesser number of threads than available cores during their execution. In such applications L2 slices which are farther to those threads are switched-off and mapped on-to L2 slices which are located nearer to those threads. By using nearer L2 slices with the help of remapped technology, some applications show improved execution time apart from reduction in leakage power consumption in NUCA caches. To estimate the maximum possible gains that can be obtained using the remap policy, we statically determine the near-optimal remap configuration using the genetic algorithms. We formulate this problem as a energy-delay product minimization problem. Our dynamic remap policy implementation gives energy-delay savings within an average of 5% than that obtained with the near-optimal remap configuration. Energy-delay product can also be minimized by improving execution time, which depends mainly on the static and dynamic NUCA access policies (DNUCA). The suitability of cache access policy depends on data sharing properties of a multi-threaded application. Hence, we propose three indices to quantify data sharing properties of an application and use them to predict a more suitable cache access policy among SNUCA and DNUCA for an application.

Network on Chips[1][2][3][4] are critical elements of modern System on Chip(SoC) as well as Chip Multiprocessor(CMP)designs. Network on Chips (NoCs) help manage high complexity of designing large chips by decoupling computation from communication. SoCs and CMPs have a multiplicity of communicating entities like programmable processing elements, hardware acceleration engines, memory blocks as well as off-chip interfaces. With power having become a serious design constraint[5], there is a great need for designing NoC which meets the target communication requirements, while minimizing power using all the tricks available at the architecture, microarchitecture and circuit levels of the de-sign. This thesis presents a holistic, QoS based, power optimal design solution of a NoC inside a CMP taking into account link microarchitecture and processor tile configurations. Guaranteeing QoS by NoCs involves guaranteeing bandwidth and throughput for connections and deterministic latencies in communication paths. Label Switching based Network-on-Chip(LS-NoC) uses a centralized LS-NoC Management framework that engineers traffic into QoS guaranteed routes. LS-NoC uses label switching, enables band-width reservation, allows physical link sharing and leverages advantages of both packet and circuit switching techniques. A flow identification algorithm takes into account band-width available in individual links to establish QoS guaranteed routes. LS-NoC caters to the requirements of streaming applications where communication channels are fixed over the lifetime of the application. The proposed NoC framework inherently supports heterogeneous and ad-hoc SoC designs. A multicast, broadcast capable label switched router for the LS-NoC has been de-signed, verified, synthesized, placed and routed and timing analyzed. A 5 port, 256 bit data bus, 4 bit label router occupies 0.431 mm2 in 130nm and delivers peak band-width of80Gbits/s per link at312.5MHz. LS Router is estimated to consume 43.08 mW. Bandwidth and latency guarantees of LS-NoC have been demonstrated on streaming applications like Hiper LAN/2 and Object Recognition Processor, Constant Bit Rate traffic patterns and video decoder traffic representing Variable Bit Rate traffic. LS-NoC was found to have a competitive figure of merit with state-of-the-art NoCs providing QoS. We envision the use of LS-NoC in general purpose CMPs where applications demand deterministic latencies and hard bandwidth requirements. Design variables for interconnect exploration include wire width, wire spacing, repeater size and spacing, degree of pipelining, supply, threshold voltage, activity and coupling factors. An optimal link configuration in terms of number of pipeline stages for a given length of link and desired operating frequency is arrived at. Optimal configurations of all links in the NoC are identified and a power-performance optimal NoC is presented. We presents a latency, power and performance trade-off study of NoCs using link microarchitecture exploration. The design and implementation of a framework for such a design space exploration study is also presented. We present the trade-off study on NoCs by varying microarchitectural(e.g. pipelining) and circuit level(e.g. frequency and voltage) parameters. A System-C based NoC exploration framework is used to explore impacts of various architectural and microarchitectural level parameters of NoC elements on power and performance of the NoC. The framework enables the designer to choose from a variety of architectural options like topology, routing policy, etc., as well as allows experimentation with various microarchitectural options for the individual links like length, wire width, pitch, pipelining, supply voltage and frequency. The framework also supports a flexible traffic generation and communication model. Latency, power and throughput results using this framework to study a 4x4 CMP are presented. The framework is used to study NoC designs of a CMP using different classes of parallel computing benchmarks[6]. One of the key findings is that the average latency of a link can be reduced by increasing pipeline depth to a certain extent, as it enables link operation at higher link frequencies. Abstract There exists an optimum degree of pipelining which minimizes the energy-delay product of the link. In a 2D Torus when the longest link is pipelined by 4 stages at which point least latency(1.56 times minimum) is achieved and power(40% of max) and throughput (64%of max) are nominal. Using frequency scaling experiments, power variations of up to40%,26.6% and24% can be seen in 2D Torus, Reduced 2D Torus and Tree based NoC between various pipeline configurations to achieve same frequency at constant voltages. Also in some cases, we find that switching to a higher pipelining configuration can actually help reduce power as the links can be designed with smaller repeaters. We also find that the overall performance of the ICNs is determined by the lengths of the links needed to support the communication patterns. Thus the mesh seems to perform the best amongst the three topologies(Mesh, Torus and Folded Torus) considered in case studies. The effects of communication overheads on performance, power and energy of a multiprocessor chip using L1,L2 cache sizes as primary exploration parameters using accurate interconnect, processor, on-chip and off-chip memory modelling are presented. On-chip and off-chip communication times have significant impact on execution time and the energy efficiency of CMPs. Large cache simply larger tile area that result in longer inter-tile communication link lengths and latencies, thus adversely impacting communication time. Smaller caches potentially have higher number of misses and frequent of off-tile communication. Energy efficient tile design is a configuration exploration and trade-off study using different cache sizes and tile areas to identify a power-performance optimal configuration for the CMP. Trade-offs are explored using a detailed, cycle accurate, multicore simulation frame-work which includes superscalar processor cores, cache coherent memory hierarchies, on-chip point-to-point communication networks and detailed interconnect model including pipelining and latency. Sapphire, a detailed multiprocessor execution environment integrating SESC, Ruby and DRAM Sim was used to run applications from the Splash2 benchmark(64KpointFFT).Link latencies are estimated for a16 core CMP simulation on Sapphire. Each tile has a single processor, L1 and L2 caches and a router. Different sizesofL1 andL2lead to different tile clock speeds, tile miss rates and tile area and hence interconnect latency. Simulations across various L1, L2 sizes indicate that the tile configuration that maximizes energy efficiency is related to minimizing communication time. Experiments also indicate different optimal tile configurations for performance, energy and energy efficiency. Clustered interconnection network, communication aware cache bank mapping and thread mapping to physical cores are also explored as potential energy saving solutions. Results indicate that ignoring link latencies can lead to large errors in estimates of program completion times, of up to 17%. Performance optimal configurations are achieved at lower L1 caches and at moderateL2 cache sizes due to higher operating frequencies and smaller link lengths and comparatively lesser communication. Using minimal L1 cache size to operate at the highest frequency may not always be the performance-power optimal choice. Larger L1 sizes, despite a drop in frequency, offer a energy advantage due to lesser communication due to misses. Clustered tile placement experiments for FFT show considerable performance per watt improvement (1.2%). Remapping most accessed L2 banks by a process in the same core or neighbouring cores after communication traffic analysis offers power and performance advantages. Remapped processes and banks in clustered tile placement show a performance per watt improvement of5.25% and energy reductionof2.53%. This suggests that processors could execute a program in multiple modes, for example, minimum energy, maximum performance.