Yong Guo, A. Varbanescu, A. Iosup, Claudio Martella, Theodore L. Willke
Processing graphs, especially at large scale, is an increasingly useful activity in a variety of business, engineering, and scientific domains. Already, there are tens of graph-processing platforms, such as Hadoop, Giraph, GraphLab, etc., each with a different design and functionality. For graph-processing to continue to evolve, users have to find it easy to select a graph-processing platform, and developers and system integrators have to find it easy to quantify the performance and other non-functional aspects of interest. However, the state of performance analysis of graph-processing platforms is still immature: there are few studies and, for the few that exist, there are few similarities, and relatively little understanding of the impact of dataset and algorithm diversity on performance. Our vision is to develop, with the help of the performance-savvy community, a comprehensive benchmarking suite for graph-processing platforms. In this work, we take a step in this direction, by proposing a set of seven challenges, summarizing our previous work on performance evaluation of distributed graph-processing platforms, and introducing our on-going work within the SPEC Research Group's Cloud Working Group.
{"title":"Benchmarking graph-processing platforms: a vision","authors":"Yong Guo, A. Varbanescu, A. Iosup, Claudio Martella, Theodore L. Willke","doi":"10.1145/2568088.2576761","DOIUrl":"https://doi.org/10.1145/2568088.2576761","url":null,"abstract":"Processing graphs, especially at large scale, is an increasingly useful activity in a variety of business, engineering, and scientific domains. Already, there are tens of graph-processing platforms, such as Hadoop, Giraph, GraphLab, etc., each with a different design and functionality. For graph-processing to continue to evolve, users have to find it easy to select a graph-processing platform, and developers and system integrators have to find it easy to quantify the performance and other non-functional aspects of interest. However, the state of performance analysis of graph-processing platforms is still immature: there are few studies and, for the few that exist, there are few similarities, and relatively little understanding of the impact of dataset and algorithm diversity on performance. Our vision is to develop, with the help of the performance-savvy community, a comprehensive benchmarking suite for graph-processing platforms. In this work, we take a step in this direction, by proposing a set of seven challenges, summarizing our previous work on performance evaluation of distributed graph-processing platforms, and introducing our on-going work within the SPEC Research Group's Cloud Working Group.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133615644","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. Pham, Victor Dogaru, R. Wagle, C. Venkatramani, Z. Kalbarczyk, R. Iyer
ZooKeeper provides scalable, highly available coordination services for distributed applications. In this paper, we evaluate the use of ZooKeeper in a distributed stream computing system called System S to provide a resilient name service, dynamic configuration management, and system state management. The evaluation shed light on the advantages of using ZooKeeper in these contexts as well as its limitations. We also describe design changes we made to handle named objects in System S to overcome the limitations. We present detailed experimental results, which we believe will be beneficial to the community.
{"title":"An evaluation of zookeeper for high availability in system S","authors":"C. Pham, Victor Dogaru, R. Wagle, C. Venkatramani, Z. Kalbarczyk, R. Iyer","doi":"10.1145/2568088.2576801","DOIUrl":"https://doi.org/10.1145/2568088.2576801","url":null,"abstract":"ZooKeeper provides scalable, highly available coordination services for distributed applications. In this paper, we evaluate the use of ZooKeeper in a distributed stream computing system called System S to provide a resilient name service, dynamic configuration management, and system state management. The evaluation shed light on the advantages of using ZooKeeper in these contexts as well as its limitations. We also describe design changes we made to handle named objects in System S to overcome the limitations. We present detailed experimental results, which we believe will be beneficial to the community.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133744571","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}
Using modelling to predict the performance characteristics of software applications typically uses Queueing Network Models representing the various system hardware resources. Leaving out the software resources, such as the limited number of threads, in such models leads to a reduced prediction accuracy. Accounting for Software Contention is a challenging task as existing techniques to model software components are complex and require deep knowledge of the software architecture. Furthermore, they also require complex measurement processes to obtain the model's service demands. In addition, solving the resultant model usually require simulation solvers which are often time consuming. In this work, we aim to provide a simpler model for three-tier web software systems which accounts for Software Contention that can be solved by time efficient analytical solvers. We achieve this by expanding the existing "Two-Level Iterative Queuing Modelling of Software Contention" method to handle the number of threads at the Application Server tier and the number of Data Sources at the Database Server tier. This is done in a generic manner to allow for extending the solution to other software components like memory and critical sections. Initial results show that our technique clearly outperforms existing techniques.
{"title":"Software contention aware queueing network model of three-tier web systems","authors":"Shadi Ghaith, Miao Wang, Philip Perry, L. Murphy","doi":"10.1145/2568088.2576760","DOIUrl":"https://doi.org/10.1145/2568088.2576760","url":null,"abstract":"Using modelling to predict the performance characteristics of software applications typically uses Queueing Network Models representing the various system hardware resources. Leaving out the software resources, such as the limited number of threads, in such models leads to a reduced prediction accuracy. Accounting for Software Contention is a challenging task as existing techniques to model software components are complex and require deep knowledge of the software architecture. Furthermore, they also require complex measurement processes to obtain the model's service demands. In addition, solving the resultant model usually require simulation solvers which are often time consuming. In this work, we aim to provide a simpler model for three-tier web software systems which accounts for Software Contention that can be solved by time efficient analytical solvers. We achieve this by expanding the existing \"Two-Level Iterative Queuing Modelling of Software Contention\" method to handle the number of threads at the Application Server tier and the number of Data Sources at the Database Server tier. This is done in a generic manner to allow for extending the solution to other software components like memory and critical sections. Initial results show that our technique clearly outperforms existing techniques.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"81 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115376487","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}
Modern software systems are expected to deliver reliable performance under highly variable load intensities while at the same time making efficient use of dynamically allocated resources. Conventional benchmarking frameworks provide limited support for emulating such highly variable and dynamic load profiles and workload scenarios. Industrial benchmarks typically use workloads with constant or stepwise increasing load intensity, or they simply replay recorded workload traces. In this paper, we present LIMBO - an Eclipse-based tool for modeling variable load intensity profiles based on the Descartes Load Intensity Model as an underlying modeling formalism.
{"title":"LIMBO: a tool for modeling variable load intensities","authors":"J. V. Kistowski, N. Herbst, Samuel Kounev","doi":"10.1145/2568088.2576092","DOIUrl":"https://doi.org/10.1145/2568088.2576092","url":null,"abstract":"Modern software systems are expected to deliver reliable performance under highly variable load intensities while at the same time making efficient use of dynamically allocated resources. Conventional benchmarking frameworks provide limited support for emulating such highly variable and dynamic load profiles and workload scenarios. Industrial benchmarks typically use workloads with constant or stepwise increasing load intensity, or they simply replay recorded workload traces. In this paper, we present LIMBO - an Eclipse-based tool for modeling variable load intensity profiles based on the Descartes Load Intensity Model as an underlying modeling formalism.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"513 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123071845","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}
Alessandro Murgia, R. Tonelli, M. Marchesi, G. Concas, S. Counsell, S. Swift
A fundamental aspect of a system's performance over time is the number of faults it generates. The relationship between the software engineering concept of "coupling" (i.e., the degree of inter-connectedness of a system's components) and faults is still a research question attracting attention and a relationship with strong implications for performance; excessive coupling is generally acknowledged to contribute to fault-proneness. In this paper, we explore the relationship between faults and coupling. Two releases from each of three open-source Eclipse projects (six releases in total) were used as an empirical basis and coupling and fault data extracted from those systems. A contrasting coupling profile between fault-free and fault-prone classes was observed and this result was statistically supported. Object-oriented (OO) classes with low values of fan-in (incoming coupling) and fan-out (outgoing coupling) appeared to support fault-free classes, while classes with high fan-out supported relatively fault-prone classes. We also considered size as an influence on fault-proneness. The study thus emphasizes the importance of minimizing coupling where possible (and particularly that of fan-out); failing to control coupling may store up problems for later in a system's life; equally, controlling class size should be a concomitant goal.
{"title":"System performance analyses through object-oriented fault and coupling prisms","authors":"Alessandro Murgia, R. Tonelli, M. Marchesi, G. Concas, S. Counsell, S. Swift","doi":"10.1145/2568088.2568089","DOIUrl":"https://doi.org/10.1145/2568088.2568089","url":null,"abstract":"A fundamental aspect of a system's performance over time is the number of faults it generates. The relationship between the software engineering concept of \"coupling\" (i.e., the degree of inter-connectedness of a system's components) and faults is still a research question attracting attention and a relationship with strong implications for performance; excessive coupling is generally acknowledged to contribute to fault-proneness. In this paper, we explore the relationship between faults and coupling. Two releases from each of three open-source Eclipse projects (six releases in total) were used as an empirical basis and coupling and fault data extracted from those systems. A contrasting coupling profile between fault-free and fault-prone classes was observed and this result was statistically supported. Object-oriented (OO) classes with low values of fan-in (incoming coupling) and fan-out (outgoing coupling) appeared to support fault-free classes, while classes with high fan-out supported relatively fault-prone classes. We also considered size as an influence on fault-proneness. The study thus emphasizes the importance of minimizing coupling where possible (and particularly that of fan-out); failing to control coupling may store up problems for later in a system's life; equally, controlling class size should be a concomitant goal.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"109 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134345158","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}
Jesús Omana Iglesias, Philip Perry, L. Murphy, Teodora Sandra Buda, James Thorburn
omputing servers generally have a narrow dynamic power range. For instance, even completely idle servers consume between 50% and 70% of their peak power. Since the usage rate of the server has the main influence on its power consumption, energy-efficiency is achieved whenever the utilization of the servers that are powered on reaches its peak. For this purpose, enterprises generally adopt the following technique: consolidate as many workloads as possible via virtualization in a minimum amount of servers (i.e. maximize utilization) and power down the ones that remain idle (i.e. reduce power consumption). However, such approach can severely impact servers' performance and reliability. In this paper, we propose a methodology to determine the ideal values for power consumption and utilization for a server without performance degradation. We accomplish this through a series of experiments using two typical types of workloads commonly found in enterprises: TPC-H and SPECpower ssj2008 benchmarks. We use the first to measure the amount of queries responded successfully per hour for different numbers of users (i.e. Throughput@Size) in the VM. Moreover, we use the latter to measure the power consumption and number of operations successfully handled by a VM at different target loads. We conducted experiments varying the utilization level and number of users for different VMs and the results show that it is possible to reach the maximum value of power consumption for a server, without experiencing performance degradations when running indi- vidual, or mixing workloads.
{"title":"An experimental methodology to evaluate energy efficiency and performance in an enterprise virtualized environment","authors":"Jesús Omana Iglesias, Philip Perry, L. Murphy, Teodora Sandra Buda, James Thorburn","doi":"10.1145/2568088.2568099","DOIUrl":"https://doi.org/10.1145/2568088.2568099","url":null,"abstract":"omputing servers generally have a narrow dynamic power range. For instance, even completely idle servers consume between 50% and 70% of their peak power. Since the usage rate of the server has the main influence on its power consumption, energy-efficiency is achieved whenever the utilization of the servers that are powered on reaches its peak. For this purpose, enterprises generally adopt the following technique: consolidate as many workloads as possible via virtualization in a minimum amount of servers (i.e. maximize utilization) and power down the ones that remain idle (i.e. reduce power consumption). However, such approach can severely impact servers' performance and reliability. In this paper, we propose a methodology to determine the ideal values for power consumption and utilization for a server without performance degradation. We accomplish this through a series of experiments using two typical types of workloads commonly found in enterprises: TPC-H and SPECpower ssj2008 benchmarks. We use the first to measure the amount of queries responded successfully per hour for different numbers of users (i.e. Throughput@Size) in the VM. Moreover, we use the latter to measure the power consumption and number of operations successfully handled by a VM at different target loads. We conducted experiments varying the utilization level and number of users for different VMs and the results show that it is possible to reach the maximum value of power consumption for a server, without experiencing performance degradations when running indi- vidual, or mixing workloads.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128691889","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 an important element of the Smart Grid, Advanced Metering Infrastructure (AMI) systems have been implemented and deployed throughout the world in the past several years. An AMI system connects millions of end devices (e.g., smart meters and sensors in the residential level) with utility control centers via an efficient two-way communication infrastructure. AMI systems are able to exchange substantial meter data and control information between utilities and end devices in real-time or near real-time. The major challenge our research was to scale ABB's Meter Data Management System (MDMS) to manage data that originates from millions of smart meters. We designed a lightweight architecture capable of collect ever-increasing large amount of meter data from various metering systems, clean, analyze, and aggregate the meter data to support various smart grid applications. To meet critical high performance requirements, various concurrency processing techniques were implemented and integrated in our prototype. Our experiments showed that on average the implemented data file parser took about 42 minutes to complete parsing, cleaning, and aggregating 5.184 billion meter reads on a single machine with the hardware configuration of 12-core CPU, 32G RAM, and SSD Hard Drives. The throughput is about 7.38 billion meter reads (206.7GB data) per hour (i.e., 1811TB/year). In addition, well-designed publish/subscribe and communication infrastructures ensure the scalability and flexibility of the system.
{"title":"Speeding up processing data from millions of smart meters","authors":"Jiang Zheng, Zhao Li, A. Dagnino","doi":"10.1145/2568088.2576798","DOIUrl":"https://doi.org/10.1145/2568088.2576798","url":null,"abstract":"As an important element of the Smart Grid, Advanced Metering Infrastructure (AMI) systems have been implemented and deployed throughout the world in the past several years. An AMI system connects millions of end devices (e.g., smart meters and sensors in the residential level) with utility control centers via an efficient two-way communication infrastructure. AMI systems are able to exchange substantial meter data and control information between utilities and end devices in real-time or near real-time. The major challenge our research was to scale ABB's Meter Data Management System (MDMS) to manage data that originates from millions of smart meters. We designed a lightweight architecture capable of collect ever-increasing large amount of meter data from various metering systems, clean, analyze, and aggregate the meter data to support various smart grid applications. To meet critical high performance requirements, various concurrency processing techniques were implemented and integrated in our prototype. Our experiments showed that on average the implemented data file parser took about 42 minutes to complete parsing, cleaning, and aggregating 5.184 billion meter reads on a single machine with the hardware configuration of 12-core CPU, 32G RAM, and SSD Hard Drives. The throughput is about 7.38 billion meter reads (206.7GB data) per hour (i.e., 1811TB/year). In addition, well-designed publish/subscribe and communication infrastructures ensure the scalability and flexibility of the system.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128067863","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}
The talk will take a broad look at performance awareness, defined as the ability to observe performance and to act on the observations. The implicit question posed in the talk is what can be done to improve various aspects of performance awareness -- be it our awareness of the various performance relevant mechanisms, our awareness of the expected software performance, our ability to attain and exploit performance awareness as software developers, and our options for implementing performance aware applications.
{"title":"Performance awareness: keynote abstract","authors":"P. Tůma","doi":"10.1145/2568088.2576097","DOIUrl":"https://doi.org/10.1145/2568088.2576097","url":null,"abstract":"The talk will take a broad look at performance awareness, defined as the ability to observe performance and to act on the observations. The implicit question posed in the talk is what can be done to improve various aspects of performance awareness -- be it our awareness of the various performance relevant mechanisms, our awareness of the expected software performance, our ability to attain and exploit performance awareness as software developers, and our options for implementing performance aware applications.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"216 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133257727","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}
Garbage collection, if not tuned properly, can considerably impact application performance. Unfortunately, configuring a garbage collector is a tedious task as only few guidelines exist and tuning is often done by trial and error. We present what is, to our knowledge, the first published work on automatically tuning Java garbage collectors in a black-box manner considering all available parameters. We propose the use of iterated local search methods to automatically compute application-specific garbage collector configurations. Our experiments show that automatic tuning can reduce garbage collection time by up to 77% for a specific application and a specific workload and by 35% on average across all benchmarks (compared to the default configuration). We evaluated our approach for 3 different garbage collectors on the DaCapo and SPECjbb benchmarks, as well as on a real-world industrial application.
{"title":"The taming of the shrew: increasing performance by automatic parameter tuning for java garbage collectors","authors":"Philipp Lengauer, H. Mössenböck","doi":"10.1145/2568088.2568091","DOIUrl":"https://doi.org/10.1145/2568088.2568091","url":null,"abstract":"Garbage collection, if not tuned properly, can considerably impact application performance. Unfortunately, configuring a garbage collector is a tedious task as only few guidelines exist and tuning is often done by trial and error. We present what is, to our knowledge, the first published work on automatically tuning Java garbage collectors in a black-box manner considering all available parameters. We propose the use of iterated local search methods to automatically compute application-specific garbage collector configurations. Our experiments show that automatic tuning can reduce garbage collection time by up to 77% for a specific application and a specific workload and by 35% on average across all benchmarks (compared to the default configuration). We evaluated our approach for 3 different garbage collectors on the DaCapo and SPECjbb benchmarks, as well as on a real-world industrial application.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114224560","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}
When performance characteristics are taken into account in a software design, models can be used to identify optimal configurations of the system's parameters. Unfortunately, for realistic scenarios, the cost of the optimization is typically high, leading to computational difficulties in the exploration of large parameter spaces. This paper proposes an approach to provably exact parameter-space pruning for a class of models of large-scale software systems analyzed with fluid techniques, efficient and scalable deterministic approximations of massively parallel stochastic models. We present a result of monotonicity of fluid solutions with respect to the model parameters, and employ it in the context of optimization programs with evolutionary algorithms by discarding candidate configurations a priori, i.e., without ever solving them, whenever they are proven to give lower fitness than other configurations. An extensive numerical validation shows that this approach yields an average twofold runtime speed-up compared to a baseline optimization algorithm that does not exploit monotonicity. Furthermore, we find that the optimal configuration is within a few percent from the true one obtained by stochastic simulation, whose solution is however orders of magnitude more expensive.
{"title":"Efficient optimization of software performance models via parameter-space pruning","authors":"M. Tribastone","doi":"10.1145/2568088.2568090","DOIUrl":"https://doi.org/10.1145/2568088.2568090","url":null,"abstract":"When performance characteristics are taken into account in a software design, models can be used to identify optimal configurations of the system's parameters. Unfortunately, for realistic scenarios, the cost of the optimization is typically high, leading to computational difficulties in the exploration of large parameter spaces. This paper proposes an approach to provably exact parameter-space pruning for a class of models of large-scale software systems analyzed with fluid techniques, efficient and scalable deterministic approximations of massively parallel stochastic models. We present a result of monotonicity of fluid solutions with respect to the model parameters, and employ it in the context of optimization programs with evolutionary algorithms by discarding candidate configurations a priori, i.e., without ever solving them, whenever they are proven to give lower fitness than other configurations. An extensive numerical validation shows that this approach yields an average twofold runtime speed-up compared to a baseline optimization algorithm that does not exploit monotonicity. Furthermore, we find that the optimal configuration is within a few percent from the true one obtained by stochastic simulation, whose solution is however orders of magnitude more expensive.","PeriodicalId":243233,"journal":{"name":"Proceedings of the 5th ACM/SPEC international conference on Performance engineering","volume":"446 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2014-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115113278","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：