In response to the emergence of multicore processors, various novel and sophisticated execution models have been introduced to fully utilize these processors. One such execution model is Thread-Level Speculation (TLS), which allows potentially dependent threads to execute speculatively in parallel. While TLS offers significant performance potential for applications that are otherwise non-parallel, extracting efficient speculative threads in the presence of complex control flow and ambiguous data dependences is a real challenge. This task is further complicated by the fact that the performance of speculative threads is often architecture-dependent, input-sensitive, and exhibits phase behaviors. Thus we propose dynamic performance tuning mechanisms that determine where and how to create speculative threads at runtime. This paper describes the design, implementation, and evaluation of hardware and software support that takes advantage of runtime performance profiles to extract efficient speculative threads. In our proposed framework, speculative threads are monitored by hardware-based performance counters and their performance impact is estimated. The creation of speculative threads is adjusted based on the estimation. This paper proposes speculative threads performance estimation techniques, that are capable of correctly determining whether speculation can improve performance for loops that corresponds to 83.8% of total loop execution time across all benchmarks. This paper also examines several dynamic performance tuning policies and finds that the best tuning policy achieves an overall speedup of 36.8%on a set of benchmarks from SPEC2000 suite, which outperforms static thread management by 9.5%.
{"title":"Dynamic performance tuning for speculative threads","authors":"Yangchun Luo, Venkatesan Packirisamy, W. Hsu, Antonia Zhai, Nikhil Mungre, Ankit Tarkas","doi":"10.1145/1555754.1555812","DOIUrl":"https://doi.org/10.1145/1555754.1555812","url":null,"abstract":"In response to the emergence of multicore processors, various novel and sophisticated execution models have been introduced to fully utilize these processors. One such execution model is Thread-Level Speculation (TLS), which allows potentially dependent threads to execute speculatively in parallel. While TLS offers significant performance potential for applications that are otherwise non-parallel, extracting efficient speculative threads in the presence of complex control flow and ambiguous data dependences is a real challenge. This task is further complicated by the fact that the performance of speculative threads is often architecture-dependent, input-sensitive, and exhibits phase behaviors. Thus we propose dynamic performance tuning mechanisms that determine where and how to create speculative threads at runtime.\u0000 This paper describes the design, implementation, and evaluation of hardware and software support that takes advantage of runtime performance profiles to extract efficient speculative threads. In our proposed framework, speculative threads are monitored by hardware-based performance counters and their performance impact is estimated. The creation of speculative threads is adjusted based on the estimation. This paper proposes speculative threads performance estimation techniques, that are capable of correctly determining whether speculation can improve performance for loops that corresponds to 83.8% of total loop execution time across all benchmarks. This paper also examines several dynamic performance tuning policies and finds that the best tuning policy achieves an overall speedup of 36.8%on a set of benchmarks from SPEC2000 suite, which outperforms static thread management by 9.5%.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"8 1","pages":"462-473"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85427150","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dynamic voltage and frequency scaling (DVFS) is a commonly-used power-management scheme that dynamically adjusts power and performance to the time-varying needs of running programs. Unfortunately, conventional DVFS, relying on off-chip regulators, faces limitations in terms of temporal granularity and high costs when considered for future multi-core systems. To overcome these challenges, this paper presents thread motion (TM), a fine-grained power-management scheme for chip multiprocessors (CMPs). Instead of incurring the high cost of changing the voltage and frequency of different cores, TM enables rapid movement of threads to adapt the time-varying computing needs of running applications to a mixture of cores with fixed but different power/performance levels. Results show that for the same power budget, two voltage/frequency levels are sufficient to provide performance gains commensurate to idealized scenarios using per-core voltage control. Thread motion extends workload-based power management into the nanosecond realm and, for a given power budget, provides up to 20% better performance than coarse-grained DVFS.
{"title":"Thread motion: fine-grained power management for multi-core systems","authors":"K. Rangan, Gu-Yeon Wei, D. Brooks","doi":"10.1145/1555754.1555793","DOIUrl":"https://doi.org/10.1145/1555754.1555793","url":null,"abstract":"Dynamic voltage and frequency scaling (DVFS) is a commonly-used power-management scheme that dynamically adjusts power and performance to the time-varying needs of running programs. Unfortunately, conventional DVFS, relying on off-chip regulators, faces limitations in terms of temporal granularity and high costs when considered for future multi-core systems. To overcome these challenges, this paper presents thread motion (TM), a fine-grained power-management scheme for chip multiprocessors (CMPs). Instead of incurring the high cost of changing the voltage and frequency of different cores, TM enables rapid movement of threads to adapt the time-varying computing needs of running applications to a mixture of cores with fixed but different power/performance levels. Results show that for the same power budget, two voltage/frequency levels are sufficient to provide performance gains commensurate to idealized scenarios using per-core voltage control. Thread motion extends workload-based power management into the nanosecond realm and, for a given power budget, provides up to 20% better performance than coarse-grained DVFS.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"46 1","pages":"302-313"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81767798","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recently proposed high-radix interconnection networks [10] require global adaptive routing to achieve optimum performance. Existing direct adaptive routing methods are slow to sense congestion remote from the source router and hence misroute many packets before such congestion is detected. This paper introduces indirect global adaptive routing (IAR) in which the adaptive routing decision uses information that is not directly available at the source router. We describe four IAR routing methods: credit round trip (CRT) [10], progressive adaptive routing (PAR), piggyback routing (PB), and reservation routing (RES). We evaluate each of these methods on the dragonfly topology under both steady-state and transient loads. Our results show that PB, PAR, and CRT all achieve good performance. PB provides the best absolute performance, with 2-7% lower latency on steady-state uniform random traffic at 70% load, while PAR provides the fastest response on transient loads. We also evaluate the implementation costs of the indirect adaptive routing methods and show that PB has the lowest implementation cost requiring <1% increase in the total storage of a typical high-radix router.
{"title":"Indirect adaptive routing on large scale interconnection networks","authors":"Nan Jiang, John Kim, W. Dally","doi":"10.1145/1555754.1555783","DOIUrl":"https://doi.org/10.1145/1555754.1555783","url":null,"abstract":"Recently proposed high-radix interconnection networks [10] require global adaptive routing to achieve optimum performance. Existing direct adaptive routing methods are slow to sense congestion remote from the source router and hence misroute many packets before such congestion is detected. This paper introduces indirect global adaptive routing (IAR) in which the adaptive routing decision uses information that is not directly available at the source router. We describe four IAR routing methods: credit round trip (CRT) [10], progressive adaptive routing (PAR), piggyback routing (PB), and reservation routing (RES). We evaluate each of these methods on the dragonfly topology under both steady-state and transient loads. Our results show that PB, PAR, and CRT all achieve good performance. PB provides the best absolute performance, with 2-7% lower latency on steady-state uniform random traffic at 70% load, while PAR provides the fastest response on transient loads. We also evaluate the implementation costs of the indirect adaptive routing methods and show that PB has the lowest implementation cost requiring <1% increase in the total storage of a typical high-radix router.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"71 1","pages":"220-231"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87740054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Andrew Putnam, S. Eggers, Dave Bennett, E. Dellinger, J. Mason, Henry Styles, P. Sundararajan, Ralph Wittig
Many-cache is a memory architecture that efficiently supports caching in commercially available FPGAs. It facilitates FPGA programming for high-performance computing (HPC) developers by providing them with memory performance that is greater and power consumption that is less than their current CPU platforms, but without sacrificing their familiar, C-based programming environment. Many-cache creates multiple, multi-banked caches on top of an FGPA's small, independent memories, each targeting a particular data structure or region of memory in an application and each customized for the memory operations that access it. The caches are automatically generated from C source by the CHiMPS C-to-FPGA compiler. This paper presents the analyses and optimizations of the CHiMPS compiler that construct many-cache caches. An architectural evaluation of CHiMPS-generated FPGAs demonstrates a performance advantage of 7.8x (geometric mean) over CPU-only execution of the same source code, FPGA power usage that is on average 4.1x less, and consequently performance per watt that is also greater, by a geometric mean of 21.3x.
{"title":"Performance and power of cache-based reconfigurable computing","authors":"Andrew Putnam, S. Eggers, Dave Bennett, E. Dellinger, J. Mason, Henry Styles, P. Sundararajan, Ralph Wittig","doi":"10.1145/1555754.1555804","DOIUrl":"https://doi.org/10.1145/1555754.1555804","url":null,"abstract":"Many-cache is a memory architecture that efficiently supports caching in commercially available FPGAs. It facilitates FPGA programming for high-performance computing (HPC) developers by providing them with memory performance that is greater and power consumption that is less than their current CPU platforms, but without sacrificing their familiar, C-based programming environment.\u0000 Many-cache creates multiple, multi-banked caches on top of an FGPA's small, independent memories, each targeting a particular data structure or region of memory in an application and each customized for the memory operations that access it. The caches are automatically generated from C source by the CHiMPS C-to-FPGA compiler.\u0000 This paper presents the analyses and optimizations of the CHiMPS compiler that construct many-cache caches. An architectural evaluation of CHiMPS-generated FPGAs demonstrates a performance advantage of 7.8x (geometric mean) over CPU-only execution of the same source code, FPGA power usage that is on average 4.1x less, and consequently performance per watt that is also greater, by a geometric mean of 21.3x.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"197 1","pages":"395-405"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74504682","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As their prices decline, their storage capacities increase, and their endurance improves, NAND Flash Solid State Disks (SSD) provide an increasingly attractive alternative to Hard Disk Drives (HDD) for portable computing systems and PCs. This paper presents a study of NAND Flash SSD architectures and their management techniques, quantifying SSD performance under user-driven/PC applications in a multi-tasked environment; user activity represents typical PC workloads and includes browsing files and folders, emailing, text editing and document creation, surfing the web, listening to music and playing movies, editing large pictures, and running office applications. We find the following: (a) the real limitation to NAND Flash memory performance is not its low per-device bandwidth but its internal core interface; (b) NAND Flash memory media transfer rates do not need to scale up to those of HDDs for good performance; (c) SSD organizations that exploit concurrency at both the system and device level (e.g. RAID-like organizations and Micron-style (superblocks) improve performance significantly; and (d) these system- and device-level concurrency mechanisms are, to a significant degree, orthogonal: that is, the performance increase due to one does not come at the expense of the other, as each exploits a different facet of concurrency exhibited within the PC workload.
{"title":"The performance of PC solid-state disks (SSDs) as a function of bandwidth, concurrency, device architecture, and system organization","authors":"Cagdas Dirik, B. Jacob","doi":"10.1145/1555754.1555790","DOIUrl":"https://doi.org/10.1145/1555754.1555790","url":null,"abstract":"As their prices decline, their storage capacities increase, and their endurance improves, NAND Flash Solid State Disks (SSD) provide an increasingly attractive alternative to Hard Disk Drives (HDD) for portable computing systems and PCs. This paper presents a study of NAND Flash SSD architectures and their management techniques, quantifying SSD performance under user-driven/PC applications in a multi-tasked environment; user activity represents typical PC workloads and includes browsing files and folders, emailing, text editing and document creation, surfing the web, listening to music and playing movies, editing large pictures, and running office applications.\u0000 We find the following: (a) the real limitation to NAND Flash memory performance is not its low per-device bandwidth but its internal core interface; (b) NAND Flash memory media transfer rates do not need to scale up to those of HDDs for good performance; (c) SSD organizations that exploit concurrency at both the system and device level (e.g. RAID-like organizations and Micron-style (superblocks) improve performance significantly; and (d) these system- and device-level concurrency mechanisms are, to a significant degree, orthogonal: that is, the performance increase due to one does not come at the expense of the other, as each exploits a different facet of concurrency exhibited within the PC workload.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"7 1","pages":"279-289"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73107166","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As chip multiprocessors (CMP) become the main trend in processor development, various power and thermal management strategies have recently been proposed to optimize system performance while controlling the power or temperature of a CMP chip to stay below a constraint. The availability of per-core DVFS (dynamic voltage and frequency scaling) also makes it possible to develop advanced management strategies. However, most existing solutions rely on open-loop search or optimization with the assumption that power can be estimated accurately, while others adopt oversimplified feedback control strategies to control power and temperature separately, without any theoretical guarantees. In this paper, we propose a chip-level power control algorithm that is systematically designed based on optimal control theory. Our algorithm can precisely control the power of a CMP chip to the desired set point while maintaining the temperature of each core below a specified threshold. Furthermore, an online model estimator is designed to achieve analytical assurance of control accuracy and system stability, even in the face of significant workload variations or unpredictable chip or core variations. Empirical results on a physical testbed show that our controller outperforms two state-of-the-art control algorithms by having better SPEC benchmark performance and more precise power control. In addition, extensive simulation results demonstrate the efficacy of our algorithm for various CMP configurations.
{"title":"Temperature-constrained power control for chip multiprocessors with online model estimation","authors":"Yefu Wang, Kai Ma, Xiaorui Wang","doi":"10.1145/1555754.1555794","DOIUrl":"https://doi.org/10.1145/1555754.1555794","url":null,"abstract":"As chip multiprocessors (CMP) become the main trend in processor development, various power and thermal management strategies have recently been proposed to optimize system performance while controlling the power or temperature of a CMP chip to stay below a constraint. The availability of per-core DVFS (dynamic voltage and frequency scaling) also makes it possible to develop advanced management strategies. However, most existing solutions rely on open-loop search or optimization with the assumption that power can be estimated accurately, while others adopt oversimplified feedback control strategies to control power and temperature separately, without any theoretical guarantees. In this paper, we propose a chip-level power control algorithm that is systematically designed based on optimal control theory. Our algorithm can precisely control the power of a CMP chip to the desired set point while maintaining the temperature of each core below a specified threshold. Furthermore, an online model estimator is designed to achieve analytical assurance of control accuracy and system stability, even in the face of significant workload variations or unpredictable chip or core variations. Empirical results on a physical testbed show that our controller outperforms two state-of-the-art control algorithms by having better SPEC benchmark performance and more precise power control. In addition, extensive simulation results demonstrate the efficacy of our algorithm for various CMP configurations.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"30 1","pages":"314-324"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85538568","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
N. Hardavellas, M. Ferdman, B. Falsafi, A. Ailamaki
Increases in on-chip communication delay and the large working sets of server and scientific workloads complicate the design of the on-chip last-level cache for multicore processors. The large working sets favor a shared cache design that maximizes the aggregate cache capacity and minimizes off-chip memory requests. At the same time, the growing on-chip communication delay favors core-private caches that replicate data to minimize delays on global wires. Recent hybrid proposals offer lower average latency than conventional designs, but they address the placement requirements of only a subset of the data accessed by the application, require complex lookup and coherence mechanisms that increase latency, or fail to scale to high core counts. In this work, we observe that the cache access patterns of a range of server and scientific workloads can be classified into distinct classes, where each class is amenable to different block placement policies. Based on this observation, we propose Reactive NUCA (R-NUCA), a distributed cache design which reacts to the class of each cache access and places blocks at the appropriate location in the cache. R-NUCA cooperates with the operating system to support intelligent placement, migration, and replication without the overhead of an explicit coherence mechanism for the on-chip last-level cache. In a range of server, scientific, and multiprogrammed workloads, R-NUCA matches the performance of the best cache design for each workload, improving performance by 14% on average over competing designs and by 32% at best, while achieving performance within 5% of an ideal cache design.
{"title":"Reactive NUCA: near-optimal block placement and replication in distributed caches","authors":"N. Hardavellas, M. Ferdman, B. Falsafi, A. Ailamaki","doi":"10.1145/1555754.1555779","DOIUrl":"https://doi.org/10.1145/1555754.1555779","url":null,"abstract":"Increases in on-chip communication delay and the large working sets of server and scientific workloads complicate the design of the on-chip last-level cache for multicore processors. The large working sets favor a shared cache design that maximizes the aggregate cache capacity and minimizes off-chip memory requests. At the same time, the growing on-chip communication delay favors core-private caches that replicate data to minimize delays on global wires. Recent hybrid proposals offer lower average latency than conventional designs, but they address the placement requirements of only a subset of the data accessed by the application, require complex lookup and coherence mechanisms that increase latency, or fail to scale to high core counts.\u0000 In this work, we observe that the cache access patterns of a range of server and scientific workloads can be classified into distinct classes, where each class is amenable to different block placement policies. Based on this observation, we propose Reactive NUCA (R-NUCA), a distributed cache design which reacts to the class of each cache access and places blocks at the appropriate location in the cache. R-NUCA cooperates with the operating system to support intelligent placement, migration, and replication without the overhead of an explicit coherence mechanism for the on-chip last-level cache. In a range of server, scientific, and multiprogrammed workloads, R-NUCA matches the performance of the best cache design for each workload, improving performance by 14% on average over competing designs and by 32% at best, while achieving performance within 5% of an ideal cache design.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"3 1","pages":"184-195"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73189765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
C. Madriles, P. López, J. M. Codina, E. Gibert, Fernando Latorre, Alejandro Martínez, Raúl Martínez, Antonio González
Industry has shifted towards multi-core designs as we have hit the memory and power walls. However, single thread performance remains of paramount importance since some applications have limited thread-level parallelism (TLP), and even a small part with limited TLP impose important constraints to the global performance, as explained by Amdahl's law. In this paper we propose a novel approach for leveraging multiple cores to improve single-thread performance in a multi-core design. The proposed technique features a set of novel hardware mechanisms that support the execution of threads generated at compile time. These threads result from a fine-grain speculative decomposition of the original application and they are executed under a modified multi-core system that includes: (1) mechanisms to support multiple versions; (2) mechanisms to detect violations among threads; (3) mechanisms to reconstruct the original sequential order; and (4) mechanisms to checkpoint the architectural state and recovery to handle misspeculations. The proposed scheme outperforms previous hardware-only schemes to implement the idea of combining cores for executing single-thread applications in a multi-core design by more than 10% on average on Spec2006 for all configurations. Moreover, single-thread performance is improved by 41% on average when the proposed scheme is used on a Tiny Core, and up to 2.6x for some selected applications.
{"title":"Boosting single-thread performance in multi-core systems through fine-grain multi-threading","authors":"C. Madriles, P. López, J. M. Codina, E. Gibert, Fernando Latorre, Alejandro Martínez, Raúl Martínez, Antonio González","doi":"10.1145/1555754.1555813","DOIUrl":"https://doi.org/10.1145/1555754.1555813","url":null,"abstract":"Industry has shifted towards multi-core designs as we have hit the memory and power walls. However, single thread performance remains of paramount importance since some applications have limited thread-level parallelism (TLP), and even a small part with limited TLP impose important constraints to the global performance, as explained by Amdahl's law.\u0000 In this paper we propose a novel approach for leveraging multiple cores to improve single-thread performance in a multi-core design. The proposed technique features a set of novel hardware mechanisms that support the execution of threads generated at compile time. These threads result from a fine-grain speculative decomposition of the original application and they are executed under a modified multi-core system that includes: (1) mechanisms to support multiple versions; (2) mechanisms to detect violations among threads; (3) mechanisms to reconstruct the original sequential order; and (4) mechanisms to checkpoint the architectural state and recovery to handle misspeculations.\u0000 The proposed scheme outperforms previous hardware-only schemes to implement the idea of combining cores for executing single-thread applications in a multi-core design by more than 10% on average on Spec2006 for all configurations. Moreover, single-thread performance is improved by 41% on average when the proposed scheme is used on a Tiny Core, and up to 2.6x for some selected applications.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"12 1","pages":"474-483"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81735261","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Abts, Natalie D. Enright Jerger, John Kim, Dan Gibson, Mikko H. Lipasti
In the near term, Moore's law will continue to provide an increasing number of transistors and therefore an increasing number of on-chip cores. Limited pin bandwidth prevents the integration of a large number of memory controllers on-chip. With many cores, and few memory controllers, where to locate the memory controllers in the on-chip interconnection fabric becomes an important and as yet unexplored question. In this paper we show how the location of the memory controllers can reduce contention (hot spots) in the on-chip fabric and lower the variance in reference latency. This in turn provides predictable performance for memory-intensive applications regardless of the processing core on which a thread is scheduled. We explore the design space of on-chip fabrics to find optimal memory controller placement relative to different topologies (i.e. mesh and torus), routing algorithms, and workloads.
{"title":"Achieving predictable performance through better memory controller placement in many-core CMPs","authors":"D. Abts, Natalie D. Enright Jerger, John Kim, Dan Gibson, Mikko H. Lipasti","doi":"10.1145/1555754.1555810","DOIUrl":"https://doi.org/10.1145/1555754.1555810","url":null,"abstract":"In the near term, Moore's law will continue to provide an increasing number of transistors and therefore an increasing number of on-chip cores. Limited pin bandwidth prevents the integration of a large number of memory controllers on-chip. With many cores, and few memory controllers, where to locate the memory controllers in the on-chip interconnection fabric becomes an important and as yet unexplored question. In this paper we show how the location of the memory controllers can reduce contention (hot spots) in the on-chip fabric and lower the variance in reference latency. This in turn provides predictable performance for memory-intensive applications regardless of the processing core on which a thread is scheduled. We explore the design space of on-chip fabrics to find optimal memory controller placement relative to different topologies (i.e. mesh and torus), routing algorithms, and workloads.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"31 1","pages":"451-461"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74356656","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many multi-core processors employ a large last-level cache (LLC) shared among the multiple cores. Past research has demonstrated that sharing-oblivious cache management policies (e.g., LRU) can lead to poor performance and fairness when the multiple cores compete for the limited LLC capacity. Different memory access patterns can cause cache contention in different ways, and various techniques have been proposed to target some of these behaviors. In this work, we propose a new cache management approach that combines dynamic insertion and promotion policies to provide the benefits of cache partitioning, adaptive insertion, and capacity stealing all with a single mechanism. By handling multiple types of memory behaviors, our proposed technique outperforms techniques that target only either capacity partitioning or adaptive insertion.
{"title":"PIPP: promotion/insertion pseudo-partitioning of multi-core shared caches","authors":"Yuejian Xie, G. Loh","doi":"10.1145/1555754.1555778","DOIUrl":"https://doi.org/10.1145/1555754.1555778","url":null,"abstract":"Many multi-core processors employ a large last-level cache (LLC) shared among the multiple cores. Past research has demonstrated that sharing-oblivious cache management policies (e.g., LRU) can lead to poor performance and fairness when the multiple cores compete for the limited LLC capacity. Different memory access patterns can cause cache contention in different ways, and various techniques have been proposed to target some of these behaviors. In this work, we propose a new cache management approach that combines dynamic insertion and promotion policies to provide the benefits of cache partitioning, adaptive insertion, and capacity stealing all with a single mechanism. By handling multiple types of memory behaviors, our proposed technique outperforms techniques that target only either capacity partitioning or adaptive insertion.","PeriodicalId":91388,"journal":{"name":"Proceedings. International Symposium on Computer Architecture","volume":"32 1","pages":"174-183"},"PeriodicalIF":0.0,"publicationDate":"2009-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81767603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：