X. Vera, J. Abella, J. Carretero, Antonio González
Soft errors are an important challenge in contemporary microprocessors. Modern processors have caches and large memory arrays protected by parity or error detection and correction codes. However, today's failure rate is dominated by flip flops, latches, and the increasing sensitivity of combinational logic to particle strikes. Moreover, as Chip Multi-Processors (CMPs) become ubiquitous, meeting the FIT budget for new designs is becoming a major challenge. Solutions based on replicating threads have been explored deeply; however, their high cost in performance and energy make them unsuitable for current designs. Moreover, our studies based on a typical configuration for a modern processor show that focusing on the top 5 most vulnerable structures can provide up to 70% reduction in FIT rate. Therefore, full replication may overprotect the chip by reducing the FIT much below budget. We propose Selective Replication, a lightweight-reconfigurable mechanism that achieves a high FIT reduction by protecting the most vulnerable instructions with minimal performance and energy impact. Low performance degradation is achieved by not requiring additional issue slots and reissuing instructions only during the time window between when they are retirable and they actually retire. Coverage can be reconfigured online by replicating only a subset of the instructions (the most vulnerable ones). Instructions' vulnerability is estimated based on the area they occupy and the time they spend in the issue queue. By changing the vulnerability threshold, we can adjust the trade-off between coverage and performance loss. Results for an out-of-order processor configured similarly to Intel® Core#8482; Micro-Architecture show that our scheme can achieve over 65% FIT reduction with less than 4% performance degradation with small area and complexity overhead.
{"title":"Selective replication: A lightweight technique for soft errors","authors":"X. Vera, J. Abella, J. Carretero, Antonio González","doi":"10.1145/1658357.1658359","DOIUrl":"https://doi.org/10.1145/1658357.1658359","url":null,"abstract":"Soft errors are an important challenge in contemporary microprocessors. Modern processors have caches and large memory arrays protected by parity or error detection and correction codes. However, today's failure rate is dominated by flip flops, latches, and the increasing sensitivity of combinational logic to particle strikes. Moreover, as Chip Multi-Processors (CMPs) become ubiquitous, meeting the FIT budget for new designs is becoming a major challenge.\u0000 Solutions based on replicating threads have been explored deeply; however, their high cost in performance and energy make them unsuitable for current designs. Moreover, our studies based on a typical configuration for a modern processor show that focusing on the top 5 most vulnerable structures can provide up to 70% reduction in FIT rate. Therefore, full replication may overprotect the chip by reducing the FIT much below budget.\u0000 We propose Selective Replication, a lightweight-reconfigurable mechanism that achieves a high FIT reduction by protecting the most vulnerable instructions with minimal performance and energy impact. Low performance degradation is achieved by not requiring additional issue slots and reissuing instructions only during the time window between when they are retirable and they actually retire. Coverage can be reconfigured online by replicating only a subset of the instructions (the most vulnerable ones). Instructions' vulnerability is estimated based on the area they occupy and the time they spend in the issue queue. By changing the vulnerability threshold, we can adjust the trade-off between coverage and performance loss.\u0000 Results for an out-of-order processor configured similarly to Intel® Core#8482; Micro-Architecture show that our scheme can achieve over 65% FIT reduction with less than 4% performance degradation with small area and complexity overhead.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"40 1","pages":"8:1-8:30"},"PeriodicalIF":1.5,"publicationDate":"2009-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80311933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
M. Aguilera, A. Merchant, Mehul A. Shah, Alistair C. Veitch, C. Karamanolis
We propose a new paradigm for building scalable distributed systems. Our approach does not require dealing with message-passing protocols, a major complication in existing distributed systems. Instead, developers just design and manipulate data structures within our service called Sinfonia. Sinfonia keeps data for applications on a set of memory nodes, each exporting a linear address space. At the core of Sinfonia is a new minitransaction primitive that enables efficient and consistent access to data, while hiding the complexities that arise from concurrency and failures. Using Sinfonia, we implemented two very different and complex applications in a few months: a cluster file system and a group communication service. Our implementations perform well and scale to hundreds of machines.
{"title":"Sinfonia: A new paradigm for building scalable distributed systems","authors":"M. Aguilera, A. Merchant, Mehul A. Shah, Alistair C. Veitch, C. Karamanolis","doi":"10.1145/1629087.1629088","DOIUrl":"https://doi.org/10.1145/1629087.1629088","url":null,"abstract":"We propose a new paradigm for building scalable distributed systems. Our approach does not require dealing with message-passing protocols, a major complication in existing distributed systems. Instead, developers just design and manipulate data structures within our service called Sinfonia. Sinfonia keeps data for applications on a set of memory nodes, each exporting a linear address space. At the core of Sinfonia is a new minitransaction primitive that enables efficient and consistent access to data, while hiding the complexities that arise from concurrency and failures. Using Sinfonia, we implemented two very different and complex applications in a few months: a cluster file system and a group communication service. Our implementations perform well and scale to hundreds of machines.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"19 1","pages":"5:1-5:48"},"PeriodicalIF":1.5,"publicationDate":"2009-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81413065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
L. Cherkasova, K. Ozonat, N. Mi, J. Symons, E. Smirni
Automated tools for understanding application behavior and its changes during the application lifecycle are essential for many performance analysis and debugging tasks. Application performance issues have an immediate impact on customer experience and satisfaction. A sudden slowdown of enterprise-wide application can effect a large population of customers, lead to delayed projects, and ultimately can result in company financial loss. Significantly shortened time between new software releases further exacerbates the problem of thoroughly evaluating the performance of an updated application. Our thesis is that online performance modeling should be a part of routine application monitoring. Early, informative warnings on significant changes in application performance should help service providers to timely identify and prevent performance problems and their negative impact on the service. We propose a novel framework for automated anomaly detection and application change analysis. It is based on integration of two complementary techniques: (i) a regression-based transaction model that reflects a resource consumption model of the application, and (ii) an application performance signature that provides a compact model of runtime behavior of the application. The proposed integrated framework provides a simple and powerful solution for anomaly detection and analysis of essential performance changes in application behavior. An additional benefit of the proposed approach is its simplicity: It is not intrusive and is based on monitoring data that is typically available in enterprise production environments. The introduced solution further enables the automation of capacity planning and resource provisioning tasks of multitier applications in rapidly evolving IT environments.
{"title":"Automated anomaly detection and performance modeling of enterprise applications","authors":"L. Cherkasova, K. Ozonat, N. Mi, J. Symons, E. Smirni","doi":"10.1145/1629087.1629089","DOIUrl":"https://doi.org/10.1145/1629087.1629089","url":null,"abstract":"Automated tools for understanding application behavior and its changes during the application lifecycle are essential for many performance analysis and debugging tasks. Application performance issues have an immediate impact on customer experience and satisfaction. A sudden slowdown of enterprise-wide application can effect a large population of customers, lead to delayed projects, and ultimately can result in company financial loss. Significantly shortened time between new software releases further exacerbates the problem of thoroughly evaluating the performance of an updated application. Our thesis is that online performance modeling should be a part of routine application monitoring. Early, informative warnings on significant changes in application performance should help service providers to timely identify and prevent performance problems and their negative impact on the service. We propose a novel framework for automated anomaly detection and application change analysis. It is based on integration of two complementary techniques: (i) a regression-based transaction model that reflects a resource consumption model of the application, and (ii) an application performance signature that provides a compact model of runtime behavior of the application. The proposed integrated framework provides a simple and powerful solution for anomaly detection and analysis of essential performance changes in application behavior. An additional benefit of the proposed approach is its simplicity: It is not intrusive and is based on monitoring data that is typically available in enterprise production environments. The introduced solution further enables the automation of capacity planning and resource provisioning tasks of multitier applications in rapidly evolving IT environments.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"38 1","pages":"6:1-6:32"},"PeriodicalIF":1.5,"publicationDate":"2009-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88196143","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Zagorodnov, K. Marzullo, L. Alvisi, T. Bressoud
This article describes an architecture that allows a replicated service to survive crashes without breaking its TCP connections. Our approach does not require modifications to the TCP protocol, to the operating system on the server, or to any of the software running on the clients. Furthermore, it runs on commodity hardware. We compare two implementations of this architecture (one based on primary/backup replication and another based on message logging) focusing on scalability, failover time, and application transparency. We evaluate three types of services: a file server, a Web server, and a multimedia streaming server. Our experiments suggest that the approach incurs low overhead on throughput, scales well as the number of clients increases, and allows recovery of the service in near-optimal time.
{"title":"Practical and low-overhead masking of failures of TCP-based servers","authors":"D. Zagorodnov, K. Marzullo, L. Alvisi, T. Bressoud","doi":"10.1145/1534909.1534911","DOIUrl":"https://doi.org/10.1145/1534909.1534911","url":null,"abstract":"This article describes an architecture that allows a replicated service to survive crashes without breaking its TCP connections. Our approach does not require modifications to the TCP protocol, to the operating system on the server, or to any of the software running on the clients. Furthermore, it runs on commodity hardware. We compare two implementations of this architecture (one based on primary/backup replication and another based on message logging) focusing on scalability, failover time, and application transparency. We evaluate three types of services: a file server, a Web server, and a multimedia streaming server. Our experiments suggest that the approach incurs low overhead on throughput, scales well as the number of clients increases, and allows recovery of the service in near-optimal time.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"87 3","pages":"4:1-4:39"},"PeriodicalIF":1.5,"publicationDate":"2009-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1145/1534909.1534911","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72391113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Counting items in a distributed system, and estimating the cardinality of multisets in particular, is important for a large variety of applications and a fundamental building block for emerging Internet-scale information systems. Examples of such applications range from optimizing query access plans in peer-to-peer data sharing, to computing the significance (rank/score) of data items in distributed information retrieval. The general formal problem addressed in this article is computing the network-wide distinct number of items with some property (e.g., distinct files with file name containing “spiderman”) where each node in the network holds an arbitrary subset, possibly overlapping the subsets of other nodes. The key requirements that a viable approach must satisfy are: (1) scalability towards very large network size, (2) efficiency regarding messaging overhead, (3) load balance of storage and access, (4) accuracy of the cardinality estimation, and (5) simplicity and easy integration in applications. This article contributes the DHS (Distributed Hash Sketches) method for this problem setting: a distributed, scalable, efficient, and accurate multiset cardinality estimator. DHS is based on hash sketches for probabilistic counting, but distributes the bits of each counter across network nodes in a judicious manner based on principles of Distributed Hash Tables, paying careful attention to fast access and aggregation as well as update costs. The article discusses various design choices, exhibiting tunable trade-offs between estimation accuracy, hop-count efficiency, and load distribution fairness. We further contribute a full-fledged, publicly available, open-source implementation of all our methods, and a comprehensive experimental evaluation for various settings.
{"title":"Distributed hash sketches: Scalable, efficient, and accurate cardinality estimation for distributed multisets","authors":"Nikos Ntarmos, P. Triantafillou, G. Weikum","doi":"10.1145/1482619.1482621","DOIUrl":"https://doi.org/10.1145/1482619.1482621","url":null,"abstract":"Counting items in a distributed system, and estimating the cardinality of multisets in particular, is important for a large variety of applications and a fundamental building block for emerging Internet-scale information systems. Examples of such applications range from optimizing query access plans in peer-to-peer data sharing, to computing the significance (rank/score) of data items in distributed information retrieval. The general formal problem addressed in this article is computing the network-wide distinct number of items with some property (e.g., distinct files with file name containing “spiderman”) where each node in the network holds an arbitrary subset, possibly overlapping the subsets of other nodes. The key requirements that a viable approach must satisfy are: (1) scalability towards very large network size, (2) efficiency regarding messaging overhead, (3) load balance of storage and access, (4) accuracy of the cardinality estimation, and (5) simplicity and easy integration in applications. This article contributes the DHS (Distributed Hash Sketches) method for this problem setting: a distributed, scalable, efficient, and accurate multiset cardinality estimator. DHS is based on hash sketches for probabilistic counting, but distributes the bits of each counter across network nodes in a judicious manner based on principles of Distributed Hash Tables, paying careful attention to fast access and aggregation as well as update costs. The article discusses various design choices, exhibiting tunable trade-offs between estimation accuracy, hop-count efficiency, and load distribution fairness. We further contribute a full-fledged, publicly available, open-source implementation of all our methods, and a comprehensive experimental evaluation for various settings.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"30 1","pages":"2:1-2:53"},"PeriodicalIF":1.5,"publicationDate":"2009-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90480440","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The key to high performance in Simultaneous MultiThreaded (SMT) processors lies in optimizing the distribution of shared resources to active threads. Existing resource distribution techniques optimize performance only indirectly. They infer potential performance bottlenecks by observing indicators, like instruction occupancy or cache miss counts, and take actions to try to alleviate them. While the corrective actions are designed to improve performance, their actual performance impact is not known since end performance is never monitored. Consequently, potential performance gains are lost whenever the corrective actions do not effectively address the actual bottlenecks occurring in the pipeline. We propose a different approach to SMT resource distribution that optimizes end performance directly. Our approach observes the impact that resource distribution decisions have on performance at runtime, and feeds this information back to the resource distribution mechanisms to improve future decisions. By evaluating many different resource distributions, our approach tries to learn the best distribution over time. Because we perform learning online, learning time is crucial. We develop a hill-climbing algorithm that quickly learns the best distribution of resources by following the performance gradient within the resource distribution space. We also develop several ideal learning algorithms to enable deeper insights through limit studies. This article conducts an in-depth investigation of hill-climbing SMT resource distribution using a comprehensive suite of 63 multiprogrammed workloads. Our results show hill-climbing outperforms ICOUNT, FLUSH, and DCRA (three existing SMT techniques) by 11.4%, 11.5%, and 2.8%, respectively, under the weighted IPC metric. A limit study conducted using our ideal learning algorithms shows our approach can potentially outperform the same techniques by 19.2%, 18.0%, and 7.6%, respectively, thus demonstrating additional room exists for further improvement. Using our ideal algorithms, we also identify three bottlenecks that limit online learning speed: local maxima, phased behavior, and interepoch jitter. We define metrics to quantify these learning bottlenecks, and characterize the extent to which they occur in our workloads. Finally, we conduct a sensitivity study, and investigate several extensions to improve our hill-climbing technique.
{"title":"Hill-climbing SMT processor resource distribution","authors":"Seungryul Choi, D. Yeung","doi":"10.1145/1482619.1482620","DOIUrl":"https://doi.org/10.1145/1482619.1482620","url":null,"abstract":"The key to high performance in Simultaneous MultiThreaded (SMT) processors lies in optimizing the distribution of shared resources to active threads. Existing resource distribution techniques optimize performance only indirectly. They infer potential performance bottlenecks by observing indicators, like instruction occupancy or cache miss counts, and take actions to try to alleviate them. While the corrective actions are designed to improve performance, their actual performance impact is not known since end performance is never monitored. Consequently, potential performance gains are lost whenever the corrective actions do not effectively address the actual bottlenecks occurring in the pipeline.\u0000 We propose a different approach to SMT resource distribution that optimizes end performance directly. Our approach observes the impact that resource distribution decisions have on performance at runtime, and feeds this information back to the resource distribution mechanisms to improve future decisions. By evaluating many different resource distributions, our approach tries to learn the best distribution over time. Because we perform learning online, learning time is crucial. We develop a hill-climbing algorithm that quickly learns the best distribution of resources by following the performance gradient within the resource distribution space. We also develop several ideal learning algorithms to enable deeper insights through limit studies.\u0000 This article conducts an in-depth investigation of hill-climbing SMT resource distribution using a comprehensive suite of 63 multiprogrammed workloads. Our results show hill-climbing outperforms ICOUNT, FLUSH, and DCRA (three existing SMT techniques) by 11.4%, 11.5%, and 2.8%, respectively, under the weighted IPC metric. A limit study conducted using our ideal learning algorithms shows our approach can potentially outperform the same techniques by 19.2%, 18.0%, and 7.6%, respectively, thus demonstrating additional room exists for further improvement. Using our ideal algorithms, we also identify three bottlenecks that limit online learning speed: local maxima, phased behavior, and interepoch jitter. We define metrics to quantify these learning bottlenecks, and characterize the extent to which they occur in our workloads. Finally, we conduct a sensitivity study, and investigate several extensions to improve our hill-climbing technique.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"22 1","pages":"1:1-1:47"},"PeriodicalIF":1.5,"publicationDate":"2009-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85472037","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Y. Qiao, F. Bustamante, P. Dinda, S. Birrer, Dong Lu
We show how to significantly improve the mean response time seen by both uploaders and downloaders in peer-to-peer data-sharing systems. Our work is motivated by the observation that response times are largely determined by the performance of the peers serving the requested objects, that is, by the peers in their capacity as servers. With this in mind, we take a close look at this server side of peers, characterizing its workload by collecting and examining an extensive set of traces. Using trace-driven simulation, we demonstrate the promise and potential problems with scheduling policies based on shortest-remaining-processing-time (SRPT), the algorithm known to be optimal for minimizing mean response time. The key challenge to using SRPT in this context is determining request service times. In addressing this challenge, we introduce two new estimators that enable predictive SRPT scheduling policies that closely approach the performance of ideal SRPT. We evaluate our approach through extensive single-server and system-level simulation coupled with real Internet deployment and experimentation.
{"title":"Improving peer-to-peer performance through server-side scheduling","authors":"Y. Qiao, F. Bustamante, P. Dinda, S. Birrer, Dong Lu","doi":"10.1145/1455258.1455260","DOIUrl":"https://doi.org/10.1145/1455258.1455260","url":null,"abstract":"We show how to significantly improve the mean response time seen by both uploaders and downloaders in peer-to-peer data-sharing systems. Our work is motivated by the observation that response times are largely determined by the performance of the peers serving the requested objects, that is, by the peers in their capacity as servers. With this in mind, we take a close look at this server side of peers, characterizing its workload by collecting and examining an extensive set of traces. Using trace-driven simulation, we demonstrate the promise and potential problems with scheduling policies based on shortest-remaining-processing-time (SRPT), the algorithm known to be optimal for minimizing mean response time. The key challenge to using SRPT in this context is determining request service times. In addressing this challenge, we introduce two new estimators that enable predictive SRPT scheduling policies that closely approach the performance of ideal SRPT. We evaluate our approach through extensive single-server and system-level simulation coupled with real Internet deployment and experimentation.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"101 3","pages":"10:1-10:30"},"PeriodicalIF":1.5,"publicationDate":"2008-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72593255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Multiprocessor scheduling in a shared multiprogramming environment can be structured as two-level scheduling, where a kernel-level job scheduler allots processors to jobs and a user-level thread scheduler schedules the work of a job on its allotted processors. We present a randomized work-stealing thread scheduler for fork-join multithreaded jobs that provides continual parallelism feedback to the job scheduler in the form of requests for processors. Our A-STEAL algorithm is appropriate for large parallel servers where many jobs share a common multiprocessor resource and in which the number of processors available to a particular job may vary during the job's execution. Assuming that the job scheduler never allots a job more processors than requested by the job's thread scheduler, A-STEAL guarantees that the job completes in near-optimal time while utilizing at least a constant fraction of the allotted processors. We model the job scheduler as the thread scheduler's adversary, challenging the thread scheduler to be robust to the operating environment as well as to the job scheduler's administrative policies. For example, the job scheduler might make a large number of processors available exactly when the job has little use for them. To analyze the performance of our adaptive thread scheduler under this stringent adversarial assumption, we introduce a new technique called trim analysis, which allows us to prove that our thread scheduler performs poorly on no more than a small number of time steps, exhibiting near-optimal behavior on the vast majority. More precisely, suppose that a job has work T1 and span T∞. On a machine with P processors, A-STEAL completes the job in an expected duration of O(T1/&Ptilde; + T∞ + L lg P) time steps, where L is the length of a scheduling quantum, and &Ptilde; denotes the O(T∞ + L lg P)-trimmed availability. This quantity is the average of the processor availability over all time steps except the O(T∞ + L lg P) time steps that have the highest processor availability. When the job's parallelism dominates the trimmed availability, that is, &Ptilde; ≪ T1/T∞, the job achieves nearly perfect linear speedup. Conversely, when the trimmed mean dominates the parallelism, the asymptotic running time of the job is nearly the length of its span, which is optimal. We measured the performance of A-STEAL on a simulated multiprocessor system using synthetic workloads. For jobs with sufficient parallelism, our experiments confirm that A-STEAL provides almost perfect linear speedup across a variety of processor availability profiles. We compared A-STEAL with the ABP algorithm, an adaptive work-stealing thread scheduler developed by Arora et al. [1998] which does not employ parallelism feedback. On moderately to heavily loaded machines with large numbers of processors, A-STEAL typically completed jobs more than twice as quickly as ABP, despite being allotted the same number or fewer processors on every step, while wasting only 10%
共享多程序设计环境中的多处理器调度可以结构化为两级调度，其中内核级作业调度器将处理器分配给作业，用户级线程调度器在分配的处理器上调度作业的工作。我们为fork-join多线程作业提供了一个随机窃取工作的线程调度器，它以请求处理器的形式向作业调度器提供持续的并行性反馈。我们的a - steal算法适用于大型并行服务器，其中许多作业共享一个公共多处理器资源，并且在作业执行期间，特定作业可用的处理器数量可能会变化。假设作业调度器分配的处理器数量永远不会超过作业线程调度器所要求的数量，那么a - steal可以保证作业在接近最佳的时间内完成，同时至少利用分配的处理器的一个常数部分。我们将作业调度器建模为线程调度器的对手，要求线程调度器对操作环境以及作业调度器的管理策略具有健壮性。例如，作业调度器可能会在作业几乎不需要大量处理器的时候使它们可用。为了在这种严格的对抗性假设下分析自适应线程调度器的性能，我们引入了一种称为修剪分析的新技术，它允许我们证明线程调度器在不超过一小部分时间步上表现不佳，而在绝大多数时间步上表现出接近最佳的行为。更准确地说，假设一个作业的功T1和张成的空间T∞。在有P个处理器的机器上，a - steal在O(T1/&Ptilde)的预期持续时间内完成任务。+ T∞+ llgp)时间步长，其中L为调度量子的长度，&Ptilde;表示O(T∞+ llg P)裁剪的可用性。这个量是处理器可用性在除具有最高处理器可用性的O(T∞+ L lg P)时间步之外的所有时间步上的平均值。当作业的并行性占精简可用性的主导地位时，即&Ptilde;≪T1/T∞，可实现近乎完美的线性加速。相反，当裁剪均值占并行度的主导地位时，作业的渐近运行时间接近其跨度的长度，这是最优的。我们使用合成工作负载在模拟的多处理器系统上测量了a - steal的性能。对于具有足够并行性的作业，我们的实验证实，a - steal在各种处理器可用性配置文件中提供了几乎完美的线性加速。我们将A-STEAL与ABP算法进行了比较，ABP算法是由Arora等人[1998]开发的一种不采用并行反馈的自适应工作窃取线程调度器。在具有大量处理器的中度到重度负载机器上，尽管每一步分配的处理器数量相同或更少，但A-STEAL完成作业的速度通常是ABP的两倍以上，而浪费的处理器周期仅为ABP浪费的10%。
{"title":"Adaptive work-stealing with parallelism feedback","authors":"Kunal Agrawal, C. Leiserson, Yuxiong He, W. Hsu","doi":"10.1145/1394441.1394443","DOIUrl":"https://doi.org/10.1145/1394441.1394443","url":null,"abstract":"Multiprocessor scheduling in a shared multiprogramming environment can be structured as two-level scheduling, where a kernel-level job scheduler allots processors to jobs and a user-level thread scheduler schedules the work of a job on its allotted processors. We present a randomized work-stealing thread scheduler for fork-join multithreaded jobs that provides continual parallelism feedback to the job scheduler in the form of requests for processors. Our A-STEAL algorithm is appropriate for large parallel servers where many jobs share a common multiprocessor resource and in which the number of processors available to a particular job may vary during the job's execution. Assuming that the job scheduler never allots a job more processors than requested by the job's thread scheduler, A-STEAL guarantees that the job completes in near-optimal time while utilizing at least a constant fraction of the allotted processors.\u0000 We model the job scheduler as the thread scheduler's adversary, challenging the thread scheduler to be robust to the operating environment as well as to the job scheduler's administrative policies. For example, the job scheduler might make a large number of processors available exactly when the job has little use for them. To analyze the performance of our adaptive thread scheduler under this stringent adversarial assumption, we introduce a new technique called trim analysis, which allows us to prove that our thread scheduler performs poorly on no more than a small number of time steps, exhibiting near-optimal behavior on the vast majority.\u0000 More precisely, suppose that a job has work T1 and span T∞. On a machine with P processors, A-STEAL completes the job in an expected duration of O(T1/&Ptilde; + T∞ + L lg P) time steps, where L is the length of a scheduling quantum, and &Ptilde; denotes the O(T∞ + L lg P)-trimmed availability. This quantity is the average of the processor availability over all time steps except the O(T∞ + L lg P) time steps that have the highest processor availability. When the job's parallelism dominates the trimmed availability, that is, &Ptilde; ≪ T1/T∞, the job achieves nearly perfect linear speedup. Conversely, when the trimmed mean dominates the parallelism, the asymptotic running time of the job is nearly the length of its span, which is optimal.\u0000 We measured the performance of A-STEAL on a simulated multiprocessor system using synthetic workloads. For jobs with sufficient parallelism, our experiments confirm that A-STEAL provides almost perfect linear speedup across a variety of processor availability profiles. We compared A-STEAL with the ABP algorithm, an adaptive work-stealing thread scheduler developed by Arora et al. [1998] which does not employ parallelism feedback. On moderately to heavily loaded machines with large numbers of processors, A-STEAL typically completed jobs more than twice as quickly as ABP, despite being allotted the same number or fewer processors on every step, while wasting only 10%","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"160 38 1","pages":"7:1-7:32"},"PeriodicalIF":1.5,"publicationDate":"2008-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89128278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Traditional operating system interfaces and network protocol implementations force some system state to be kept on both sides of a connection. This state ties the connection to its endpoints, impedes transparent failover, permits denial-of-service attacks, and limits scalability. This article introduces a novel TCP-like transport protocol and a new interface to replace sockets that together enable all state to be kept on one endpoint, allowing the other endpoint, typically the server, to operate without any per-connection state. Called Trickles, this approach enables servers to scale well with increasing numbers of clients, consume fewer resources, and better resist denial-of-service attacks. Measurements on a full implementation in Linux indicate that Trickles achieves performance comparable to TCP/IP, interacts well with other flows, and scales well. Trickles also enables qualitatively different kinds of networked services. Services can be geographically replicated and contacted through an anycast primitive for improved availability and performance. Widely-deployed practices that currently have client-observable side effects, such as periodic server reboots, connection redirection, and failover, can be made transparent, and perform well, under Trickles. The protocol is secure against tampering and replay attacks, and the client interface is backward-compatible, requiring no changes to sockets-based client applications.
{"title":"A stateless approach to connection-oriented protocols","authors":"Alan Shieh, A. Myers, E. G. Sirer","doi":"10.1145/1394441.1394444","DOIUrl":"https://doi.org/10.1145/1394441.1394444","url":null,"abstract":"Traditional operating system interfaces and network protocol implementations force some system state to be kept on both sides of a connection. This state ties the connection to its endpoints, impedes transparent failover, permits denial-of-service attacks, and limits scalability. This article introduces a novel TCP-like transport protocol and a new interface to replace sockets that together enable all state to be kept on one endpoint, allowing the other endpoint, typically the server, to operate without any per-connection state. Called Trickles, this approach enables servers to scale well with increasing numbers of clients, consume fewer resources, and better resist denial-of-service attacks. Measurements on a full implementation in Linux indicate that Trickles achieves performance comparable to TCP/IP, interacts well with other flows, and scales well. Trickles also enables qualitatively different kinds of networked services. Services can be geographically replicated and contacted through an anycast primitive for improved availability and performance. Widely-deployed practices that currently have client-observable side effects, such as periodic server reboots, connection redirection, and failover, can be made transparent, and perform well, under Trickles. The protocol is secure against tampering and replay attacks, and the client interface is backward-compatible, requiring no changes to sockets-based client applications.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"9 1","pages":"8:1-8:50"},"PeriodicalIF":1.5,"publicationDate":"2008-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87129931","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This article presents RaWMS, a novel lightweight random membership service for ad hoc networks. The service provides each node with a partial uniformly chosen view of network nodes. Such a membership service is useful, for example, in data dissemination algorithms, lookup and discovery services, peer sampling services, and complete membership construction. The design of RaWMS is based on a novel reverse random walk (RW) sampling technique. The article includes a formal analysis of both the reverse RW sampling technique and RaWMS and verifies it through a detailed simulation study. In addition, RaWMS is compared both analytically and by simulations with a number of other known methods such as flooding and gossip-based techniques.
{"title":"RaWMS - Random Walk Based Lightweight Membership Service for Wireless Ad Hoc Networks","authors":"Ziv Bar-Yossef, R. Friedman, G. Kliot","doi":"10.1145/1365815.1365817","DOIUrl":"https://doi.org/10.1145/1365815.1365817","url":null,"abstract":"This article presents RaWMS, a novel lightweight random membership service for ad hoc networks. The service provides each node with a partial uniformly chosen view of network nodes. Such a membership service is useful, for example, in data dissemination algorithms, lookup and discovery services, peer sampling services, and complete membership construction. The design of RaWMS is based on a novel reverse random walk (RW) sampling technique. The article includes a formal analysis of both the reverse RW sampling technique and RaWMS and verifies it through a detailed simulation study. In addition, RaWMS is compared both analytically and by simulations with a number of other known methods such as flooding and gossip-based techniques.","PeriodicalId":50918,"journal":{"name":"ACM Transactions on Computer Systems","volume":"18 1","pages":"5:1-5:66"},"PeriodicalIF":1.5,"publicationDate":"2008-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78927223","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","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：