The Experts below are selected from a list of 252 Experts worldwide ranked by ideXlab platform
Gürhan Küçük - One of the best experts on this subject based on the ideXlab platform.
-
Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
-
HPCS - Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
Burak Sezin Ovant - One of the best experts on this subject based on the ideXlab platform.
-
Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
-
HPCS - Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
S Nandi - One of the best experts on this subject based on the ideXlab platform.
-
dark matter Parallel Universe and multiple higgs signals at the ilc
arXiv: High Energy Physics - Phenomenology, 2014Co-Authors: Shreyashi Chakdar, Kirtiman Ghosh, S NandiAbstract:The existence of dark matter is now well established by several indirect experiments. Several candidates for dark matter has also been proposed. However, the dark matter can just be like our ordinary matter in a Parallel Universe with both Universes being described by their own non-abelian gauge symmetries forbidding any kinetic mixing. However, the quartic Higgs interactions involving Higgs fields between the two Universes are allowed by the symmetries of the model. The ensuing mixing between the two lightest Standard Model like Higgses gives rise to interesting signatures at the proposed international electron-positron collider (ILC) specially in the case when mass splitting between the two surviving light Higgs bosons are small ($\sim$ 100 MeV) so that they can not be resolved at the LHC.
-
Parallel Universe dark matter and invisible higgs decays
Physics Letters B, 2014Co-Authors: Shreyashi Chakdar, Kirtiman Ghosh, S NandiAbstract:Abstract The existence of the dark matter with amount about five times the ordinary matter is now well established experimentally. There are now many candidates for this dark matter. However, dark matter could be just like the ordinary matter in a Parallel Universe. If both Universes are described by a non-abelian gauge symmetries, then there will be no kinetic mixing between the ordinary photon and the dark photon, and the dark proton, dark electron and the corresponding dark nuclei, belonging to the Parallel Universe, will be stable. If the strong coupling constant, ( α s ) dark in the Parallel Universe is five times that of α s , then the dark proton will be about five time heavier, explaining why the dark matter is five times the ordinary matter. However, the two sectors will still interact via the Higgs boson of the two sectors. This will lead to the existence of a second light Higgs boson, just like the Standard Model Higgs boson. This gives rise to the invisible decay modes of the Higgs boson which can be tested at the LHC, and the proposed ILC.
Linda Freeman - One of the best experts on this subject based on the ideXlab platform.
-
a Parallel Universe competing interpretations of zimbabwe s crisis
Journal of Contemporary African Studies, 2014Co-Authors: Linda FreemanAbstract:This article examines competing interpretations of the nature and cause of Zimbabwe’s contemporary crisis. It finds that while neoliberal macroeconomic policies promoted by international financial institutions helped to provide a structural basis for the crisis, arguments attributing blame to Britain and to wider Western sanctions are overblown and inaccurate. Similarly, although Western reactions to Zimbabwe’s land reform have had a racist tinge, these paled in comparison with the explicit racist intent of policies adopted by the Zimbabwean Government. The claim that Zimbabwe is undergoing a process of progressive transformation must be weighed against the nature of state power, the intensification of class divisions, a precipitous economic decline, a problematic development strategy and the extreme abuse of human, civil and political rights.
-
A Parallel Universe – competing interpretations of Zimbabwe's crisis
Journal of Contemporary African Studies, 2014Co-Authors: Linda FreemanAbstract:This article examines competing interpretations of the nature and cause of Zimbabwe’s contemporary crisis. It finds that while neoliberal macroeconomic policies promoted by international financial institutions helped to provide a structural basis for the crisis, arguments attributing blame to Britain and to wider Western sanctions are overblown and inaccurate. Similarly, although Western reactions to Zimbabwe’s land reform have had a racist tinge, these paled in comparison with the explicit racist intent of policies adopted by the Zimbabwean Government. The claim that Zimbabwe is undergoing a process of progressive transformation must be weighed against the nature of state power, the intensification of class divisions, a precipitous economic decline, a problematic development strategy and the extreme abuse of human, civil and political rights.
İsa Ahmet Güney - One of the best experts on this subject based on the ideXlab platform.
-
Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
-
HPCS - Allocation of last level cache partitions through thread classification with Parallel Universes
2016 International Conference on High Performance Computing & Simulation (HPCS), 2016Co-Authors: Burak Sezin Ovant, İsa Ahmet Güney, Muhammed Emin Savaş, Gürhan KüçükAbstract:Last Level Caches (LLCs) are among the most common processor resources that are shared by multiple threads in simultaneous multithreaded (SMT) and chip multi processors (CMP). In an unmanaged cache organization, threads might not get along well and steal cache lines from each other. This type of thread interaction prevents effective utilization of this precious resource. Cache partitioning is one of the well-studied methods that target improved system performance through isolation of cache lines dedicated to each thread. In this study, we propose a new allocation policy that chooses the amount of cache partitions through thread classification and auxiliary cache structures, which we call Parallel Universe Tag Directories (PUTDs). Each thread maintains a PUTD structure, which enables collecting statistics from another execution dimension, where the dedicated thread receives more cache resources. Our test results show that our proposed mechanism gives better performance and fairness results with negligible hardware requirements compared to the current state of the art, in all studied processor configurations.
