With the software applications increasing in complexity, description of hardware is becoming increasingly relevant. To ensure the quality of service for specific applications, it is imperative to have an insight into hardware resources. Cache memory is used for storing data closer to the processor needed for quick access and improves the quality of service of applications. The description of cache memory usually consists of the size of different cache levels, set associativity, or line size. Software applications would benefit more from a more detailed model of cache memory.In this thesis, we offer a way of describing the behavior of cache memory which benefits software performance. Several performance events are tested, including L1 cache misses, L2 cache misses, and L3 cache misses. With the collected information, we develop performance models of cache memory behavior. Goodness of fit is tested for these models and they are used to predict the behavior of the cache memory during future runs of the same application.Our experiments show that L1 cache misses can be modeled to predict the future runs. L2 cache misses model is less accurate but still usable for predictions, and L3 cache misses model is the least accurate and is not feasible to predict the behavior of the future runs.

Single-ISA heterogeneous multi-core processors (SHMP) have become increasingly important due to their potential to significantly improve the execution efficiency for diverse workloads and thereby alleviate the power density constraints in Chip Multiprocessors (CMP). The importance of SHMP is further underscored by the fact that manufacturing defects and process variation could also cause single-ISA heterogeneity in CMPs even though the CMP is originally designed as homogeneous. However, to fully exploit the execution efficiency that SHMP has to offer, programs have to be efficiently mapped/scheduled to the appropriate cores such that the hardware resources of the cores match the resource demands of the programs, which is challenging and remains an open problem. This dissertation presents a comprehensive set of off-line and on-line techniques that leverage analytical performance modeling to bridge the gap between the workload diversity and the hardware heterogeneity. For the off-line scenario, this dissertation presents an efficient resource demand analysis framework that can estimate the resource demands of a program based on the inherent characteristics of the program without using any detailed simulation. Based on the estimated resource demands, this dissertation further proposes a multi-dimensional program-core matching technique that projects program resource demands and core configurations to a unified multi-dimensional space, and uses the weighted Euclidean distance between these two to identify the matching program-core pair. This dissertation also presents a dynamic and predictive application scheduler for SHMPs. It uses a set of hardware-efficient online profilers and an analytical performance model to simultaneously predict the application’s performance on different cores. Based on the predicted performance, the scheduler identifies and enforces near-optimal application assignment for each scheduling interval without any trial runs or off-line profiling. Using only a few kilo-bytes of extra hardware, the proposed heterogeneity-aware scheduler improves the weighted speedup by 11.3% compared with the commodity OpenSolaris scheduler and by 6.8% compared with the best known research scheduler. Finally, this dissertation presents a predictive yet cost effective mechanism to manage intra-core and/or inter-core resources in dynamic SHMP. It also uses a set of hardware-efficient online profilers and an analytical performance model to predict the application’s performance with different resource allocations. Based on the predicted performance, the resource allocator identifies and enforces near optimum resource partitions for each epoch without any trial runs. The experimental results show that the proposed predictive resource management framework could improve the weighted speedup of the CMP system by an average of 11.6% compared with the equal partition scheme, and 9.3% compared with existing reactive resource management scheme. / text

No description available.

Optical networking has been developing for decades and wavelength-division multiplexing (WDM) is the main technology used to carry signals in fiber-optical communication systems. However, its development has slowed because it is approaching the Shannon limit of nonlinear fiber transmission. Researchers are looking for multi dimensional multiplexing. Space-division multiplexing (SDM) is an ideal way to scale network capacities. The capacity of WDM/SDM network could be expanded to several times the capacity of WDM network but the active hardware devices may also increase by several times. This project aims to answer a practical question: How to construct a WDM/SDM network with less hardware resource? There is no mature research about WDM/SDM network yet. Therefore, the problem can be divided into two parts: (1) how to build a WDM/SDM network and (2) how to allocate resource and compute routes in such a network to minimize hardware resources. First, this thesis proposes a WDM/SDM node which has bypass connections between different fibers and architecture on demand (AoD) to effectively decrease the number of active hardware devices within the node. Then, two types of networks were constructed: one with bypass connections in each node and another one without any bypass connections. These networks were under the control of a software defined network (SDN) controller. The controller knew the wavelength resources within the networks. Several algorithms were applied to these networks to evaluate the effect of a bypass network and to identify the desired characteristics (to find short length path and decrease the probability of spectrum fragmentation) of an algorithm suitable for a network with bypass connections. The results of applying the proposed algorithms in two networks proved that the bypass connections increased the blocking probability in small topology but did not affect the results in large topology. The results in a large-scale network with bypass network were almost the same as the results in a network without bypass connections. Thus, bypass connections are suitable for large-scale network. / Optiskt nätverk har utvecklats i årtionden och våglängdsdelningsmultiplexering (WDM) är den viktigaste tekniken som används för att bära signaler i fiberoptiska kommunikationssystem. Utvecklingen har dock minskat eftersom den närmar sig Shannon-gränsen för olinjär fiberöverföring. Forskare letar efter flerdimensionell multiplexering. Space-division multiplexing (SDM) är ett idealiskt sätt att skala nätverkskapacitet. Kapaciteten för WDM / SDM-nätverk kan utökas till flera gånger WDM-nätverkets kapacitet, men de aktiva hårdvaraenheterna kan också öka med flera gånger. Projektet syftar till att svara på en praktisk fråga: Hur konstruerar jag ett WDM / SDM-nätverk med mindre hårdvara? Det finns ingen mogen forskning om WDM / SDM-nätverk än. Därför kan problemet delas in i två delar: (1) hur man bygger ett WDM / SDM-nätverk och (2) hur man fördelar resurser och beräknar rutter i ett sådant nätverk för att minimera hårdvaruressurser. Först föreslår denna avhandling en WDM / SDM-nod som har förbikopplingsanslutningar mellan olika fibrer och arkitektur på begäran (AoD) för att effektivt minska antalet aktiva hårdvaraenheter inom noden. Sedan konstruerades två typer av nätverk: en med bypass-anslutningar i varje nod och en annan utan några bypass-anslutningar. Dessa nätverk kontrollerades av en mjukvarudefinierad nätverkskontroller (SDN). Styrenheten visste våglängdsresurserna i nätverket. Flera algoritmer applicerades på dessa nätverk för att utvärdera effekten av ett förbikopplingsnätverk och för att identifiera de önskade egenskaperna (för att hitta en kort längdväg och minska sannolikheten för spektrumfragmentering) av en algoritm som är lämplig för ett nätverk med bypass-anslutningar. Resultaten av att tillämpa de föreslagna algoritmerna i två nät visade att förbikopplingsförbindelserna ökade blockeringssannolikheten i liten topologi men inte påverkade resultaten i stor topologi. Resultaten i ett storskaligt nätverk med bypass-nätverk var nästan samma som resultaten i ett nätverk utan bypass-anslutningar. Bypassanslutningar är således lämpliga för storskaliga nätverk.