Pub Date : 2021-09-03DOI: 10.1080/17445760.2021.1934837
M. Burgin
In this paper, classes of automata that perform distributed computations with unconventional interaction are described and studied. These automata are called selective machines and they are more powerful than Turing machines while their high computing and recognising power can be achieved exclusively by interaction when a system of recursive algorithms (automata) becomes super-recursive due to their interaction. Computations of selective machines are described by selective algorithms, which are super-recursive allowing computations of functions that are incomputable by Turing machines. Examples of selective algorithms are grammars with prohibition, correction grammars and grammars with exclusion. The study of selective machines and selective algorithms is based on the axiomatic theory of algorithms, in which the results are obtained in the general situation of axiomatically defined classes of automata and algorithms. Then these results are specified for many concrete classes of automata and algorithms, such as finite automata or Turing machines, by checking the necessary axioms.
{"title":"Modelling distributive computation by selective machines","authors":"M. Burgin","doi":"10.1080/17445760.2021.1934837","DOIUrl":"https://doi.org/10.1080/17445760.2021.1934837","url":null,"abstract":"In this paper, classes of automata that perform distributed computations with unconventional interaction are described and studied. These automata are called selective machines and they are more powerful than Turing machines while their high computing and recognising power can be achieved exclusively by interaction when a system of recursive algorithms (automata) becomes super-recursive due to their interaction. Computations of selective machines are described by selective algorithms, which are super-recursive allowing computations of functions that are incomputable by Turing machines. Examples of selective algorithms are grammars with prohibition, correction grammars and grammars with exclusion. The study of selective machines and selective algorithms is based on the axiomatic theory of algorithms, in which the results are obtained in the general situation of axiomatically defined classes of automata and algorithms. Then these results are specified for many concrete classes of automata and algorithms, such as finite automata or Turing machines, by checking the necessary axioms.","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"395 - 411"},"PeriodicalIF":1.1,"publicationDate":"2021-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1934837","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"60344392","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}
Pub Date : 2021-08-31DOI: 10.1080/17445760.2021.1971665
A. Doroshenko, E. Tulika, O. Yatsenko
ABSTRACT The centralised orchestration technique is often a bad idea for massive parallelism applications if we want to achieve a scalable solution. In this paper for this purpose, the choreography approach is undertaken and some adaptive methods and software tools of distributed implementation are proposed to enhance computation parallelism applied to the optimisation of a class of block-recursive algorithms. A new formal model of distribution and coordination of the tasks in a computing cluster as asynchronous reactive processes with message-passing represented with an actor model and choreography of actors is developed. Also, a new scheme of data placement in a multiprocessor cluster based on prioritisation of block-recursive operations is developed to reduce idling time, data movement time. Adaptive adjustment of the data placement in a cluster at run time to account for current cluster load is developed and an auto-tuning of the actor placement in a cluster based on previous statistics for optimisation is implemented. The experiments show that the choreography of actors allows to remove the central coordinating element, to avoid hard dependencies between cluster nodes, and to achieve a better degree of the parallel applications’ scalability. GRAPHICAL ABSTRACT
{"title":"Enhancing parallelism of distributed algorithms with the actor model and a smart data movement technique","authors":"A. Doroshenko, E. Tulika, O. Yatsenko","doi":"10.1080/17445760.2021.1971665","DOIUrl":"https://doi.org/10.1080/17445760.2021.1971665","url":null,"abstract":"ABSTRACT The centralised orchestration technique is often a bad idea for massive parallelism applications if we want to achieve a scalable solution. In this paper for this purpose, the choreography approach is undertaken and some adaptive methods and software tools of distributed implementation are proposed to enhance computation parallelism applied to the optimisation of a class of block-recursive algorithms. A new formal model of distribution and coordination of the tasks in a computing cluster as asynchronous reactive processes with message-passing represented with an actor model and choreography of actors is developed. Also, a new scheme of data placement in a multiprocessor cluster based on prioritisation of block-recursive operations is developed to reduce idling time, data movement time. Adaptive adjustment of the data placement in a cluster at run time to account for current cluster load is developed and an auto-tuning of the actor placement in a cluster based on previous statistics for optimisation is implemented. The experiments show that the choreography of actors allows to remove the central coordinating element, to avoid hard dependencies between cluster nodes, and to achieve a better degree of the parallel applications’ scalability. GRAPHICAL ABSTRACT","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"565 - 578"},"PeriodicalIF":1.1,"publicationDate":"2021-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48451020","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}
Pub Date : 2021-08-22DOI: 10.1080/17445760.2021.1967350
L. T. Theu, Q. Tran, Vijender Kumar Solanki, Tatiana R. Shemeleva, Duc-Tan Tran
The greatest advantage of scattering theory-based ultrasound tomography (UT) is its ability to investigate small structures. DBIM is the Distorted Born Iterative Method. The nearest neighbour interpolation method is used to enhance the reconstruction performance and reduce the reconstruction time. The raw (N 1 × N 1) and dense (N 2 × N 2) meshed integration areas are reconstructed in NN 1 and NN 2 iterations, respectively. However, choosing the best value of NN 1 to get the highest performance was not mentioned in previous works. If it is not well chosen, the reconstruction quality is even worse than that when using no interpolation. This study proposes a method to enhance the UT reconstruction by using the nearest neighbour interpolation (MR-DBIM). The corresponding algorithms are specified by the graphical concurrent programming language of Sleptsov nets. Some significant results are (1) the MR-DBIM is only meaningful when (i.e. sparse scattering domain); (2) the best performance is obtained in the DBIM when Nt = Nr , but in the MR-DBIM when Nr = 2Nt ; (3) the well-investigated value of NN 1 is 2 when and is 3 when . GRAPHICAL ABSTRACT
基于散射理论的超声层析成像(UT)的最大优点是它能够研究小结构。DBIM是一种畸变Born迭代法。采用最近邻插值法提高了重建性能，缩短了重建时间。原始(N 1 × N 1)和密集(N 2 × N 2)网格积分区域分别在NN 1和NN 2迭代中重构。然而，选择NN 1的最佳值来获得最高的性能在之前的作品中并没有提到。如果选择不好，重建质量甚至比不使用插值时还要差。本研究提出了一种利用最近邻插值(MR-DBIM)增强UT重建的方法。相应的算法由Sleptsov网络的图形并行编程语言指定。一些重要的结果是:(1)MR-DBIM只有在(即稀疏散射域)时才有意义;(2)当Nt = Nr时，DBIM的性能最好，而当Nr = 2Nt时，MR-DBIM的性能最好;(3)当，nnn 1的值为2，当，nnn 1的值为3。图形抽象
{"title":"Influence of the multi-resolution technique on tomographic reconstruction in ultrasound tomography","authors":"L. T. Theu, Q. Tran, Vijender Kumar Solanki, Tatiana R. Shemeleva, Duc-Tan Tran","doi":"10.1080/17445760.2021.1967350","DOIUrl":"https://doi.org/10.1080/17445760.2021.1967350","url":null,"abstract":"The greatest advantage of scattering theory-based ultrasound tomography (UT) is its ability to investigate small structures. DBIM is the Distorted Born Iterative Method. The nearest neighbour interpolation method is used to enhance the reconstruction performance and reduce the reconstruction time. The raw (N 1 × N 1) and dense (N 2 × N 2) meshed integration areas are reconstructed in NN 1 and NN 2 iterations, respectively. However, choosing the best value of NN 1 to get the highest performance was not mentioned in previous works. If it is not well chosen, the reconstruction quality is even worse than that when using no interpolation. This study proposes a method to enhance the UT reconstruction by using the nearest neighbour interpolation (MR-DBIM). The corresponding algorithms are specified by the graphical concurrent programming language of Sleptsov nets. Some significant results are (1) the MR-DBIM is only meaningful when (i.e. sparse scattering domain); (2) the best performance is obtained in the DBIM when Nt = Nr , but in the MR-DBIM when Nr = 2Nt ; (3) the well-investigated value of NN 1 is 2 when and is 3 when . GRAPHICAL ABSTRACT","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"579 - 593"},"PeriodicalIF":1.1,"publicationDate":"2021-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45990088","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}
Pub Date : 2021-07-23DOI: 10.1080/17445760.2021.1955113
I. V. Stetsenko, A. A. Pavlov, Oleksandra Dyfuchyna
Parallel algorithms are problematic to develop because of the negative influence of synchronisation, complicated behaviour of threads’ capturing computing resources. Experimental results show performance time’s strong dependence on algorithm parameters, such as the number of subtasks and the complexity of each task. The optimal value of subtask complexity is revealed for the particular algorithm. It is the same for different complexity of the parallelised task (with the same computing resource). To guarantee algorithm speed-up it is important to have a method for investigating the efficiency of parallel algorithm before its implementation on specified computing resources. Stochastic Petri net potentially could be a high accuracy tool for investigating the efficiency of a parallel algorithm. However, a huge number of elements are needed to compose a model of non-trivial algorithm that limits the application of this tool in practice. Petri-object simulation method allows replication of Petri nets with specified parameters and model creation of a list of linked Petri-objects. Basic templates for the model creation of a multithreaded algorithm are developed. Applying these templates, the model of the parallel discrete event simulation algorithm is developed and investigated. By the model results, the algorithm parameters providing the least performance time can be determined.
{"title":"Parallel algorithm development and testing using Petri-object simulation","authors":"I. V. Stetsenko, A. A. Pavlov, Oleksandra Dyfuchyna","doi":"10.1080/17445760.2021.1955113","DOIUrl":"https://doi.org/10.1080/17445760.2021.1955113","url":null,"abstract":"Parallel algorithms are problematic to develop because of the negative influence of synchronisation, complicated behaviour of threads’ capturing computing resources. Experimental results show performance time’s strong dependence on algorithm parameters, such as the number of subtasks and the complexity of each task. The optimal value of subtask complexity is revealed for the particular algorithm. It is the same for different complexity of the parallelised task (with the same computing resource). To guarantee algorithm speed-up it is important to have a method for investigating the efficiency of parallel algorithm before its implementation on specified computing resources. Stochastic Petri net potentially could be a high accuracy tool for investigating the efficiency of a parallel algorithm. However, a huge number of elements are needed to compose a model of non-trivial algorithm that limits the application of this tool in practice. Petri-object simulation method allows replication of Petri nets with specified parameters and model creation of a list of linked Petri-objects. Basic templates for the model creation of a multithreaded algorithm are developed. Applying these templates, the model of the parallel discrete event simulation algorithm is developed and investigated. By the model results, the algorithm parameters providing the least performance time can be determined.","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"549 - 564"},"PeriodicalIF":1.1,"publicationDate":"2021-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1955113","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"42215247","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}
Pub Date : 2021-07-20DOI: 10.1080/17445760.2021.1952580
D. Zaitsev, T. Shmeleva, B. Pröll
Multidimensional torus interconnect finds wide application in modern exascale computing. For models design in high-performance computing, grid and cloud computing, and also systems biology, two basic ways of specifying spatial structures with Petri nets are considered – an infinite Petri net specified by a parametric expression (PE) and a reenterable coloured Petri net (CPN). The paper studies a composition of hypertorus grid models in the form of a PE and a reenterable CPN, their mutual transformations, and unfolding into a place/transition net; the parameters are the number of dimensions and the size of grid. A grid is composed via connection of neighbouring cells by dedicated transitions modelling channels. Reenterable model peculiarities are explained on step-by-step simulation examples. The rules of mutual transformations of Petri net spatial specifications are specified. Comparative investigation of two mentioned forms of spatial specifications is implemented, including analysis techniques and tools. CPNs are convenient for the state space analysis. The main advantage of PEs is the ability to obtain linear invariants and other structural constructs of Petri nets, for instance, siphons and traps, in parametric form that allows us to draw conclusions on Petri net properties for any values of parameters.
{"title":"Spatial specification of hypertorus interconnect by infinite and reenterable coloured Petri nets","authors":"D. Zaitsev, T. Shmeleva, B. Pröll","doi":"10.1080/17445760.2021.1952580","DOIUrl":"https://doi.org/10.1080/17445760.2021.1952580","url":null,"abstract":"Multidimensional torus interconnect finds wide application in modern exascale computing. For models design in high-performance computing, grid and cloud computing, and also systems biology, two basic ways of specifying spatial structures with Petri nets are considered – an infinite Petri net specified by a parametric expression (PE) and a reenterable coloured Petri net (CPN). The paper studies a composition of hypertorus grid models in the form of a PE and a reenterable CPN, their mutual transformations, and unfolding into a place/transition net; the parameters are the number of dimensions and the size of grid. A grid is composed via connection of neighbouring cells by dedicated transitions modelling channels. Reenterable model peculiarities are explained on step-by-step simulation examples. The rules of mutual transformations of Petri net spatial specifications are specified. Comparative investigation of two mentioned forms of spatial specifications is implemented, including analysis techniques and tools. CPNs are convenient for the state space analysis. The main advantage of PEs is the ability to obtain linear invariants and other structural constructs of Petri nets, for instance, siphons and traps, in parametric form that allows us to draw conclusions on Petri net properties for any values of parameters.","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"37 1","pages":"1 - 21"},"PeriodicalIF":1.1,"publicationDate":"2021-07-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1952580","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43572163","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}
Pub Date : 2021-07-14DOI: 10.1080/17445760.2021.1952579
Hongzhou Zhu, J. Meng
Let G be a connected graph, F be a subset of , S be a subset of . The cyclic vertex connectivity of G, denoted by , is the minimum cardinality of F such that G−F is disconnected and at least two of its components contain cycles. The cyclic edge connectivity of G, denoted by , is the minimum cardinality of S such that G−S is disconnected and at least two of its components contain cycles. Let denote the data center network. In this paper, we obtain the following results: for ; for , ; for ; for ; for , .
{"title":"Cyclic connectivity of the data center network","authors":"Hongzhou Zhu, J. Meng","doi":"10.1080/17445760.2021.1952579","DOIUrl":"https://doi.org/10.1080/17445760.2021.1952579","url":null,"abstract":"Let G be a connected graph, F be a subset of , S be a subset of . The cyclic vertex connectivity of G, denoted by , is the minimum cardinality of F such that G−F is disconnected and at least two of its components contain cycles. The cyclic edge connectivity of G, denoted by , is the minimum cardinality of S such that G−S is disconnected and at least two of its components contain cycles. Let denote the data center network. In this paper, we obtain the following results: for ; for , ; for ; for ; for , .","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"623 - 629"},"PeriodicalIF":1.1,"publicationDate":"2021-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1952579","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"46926444","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}
Pub Date : 2021-06-27DOI: 10.1080/17445760.2021.1945054
A. Kostikov, N. D. Zaitsev, O. Subotin
The present article is focused on the application of Sleptsov net (SN) for solving system of linear algebraic equations with a tridiagonal matrix. SN, which implements the double sweep algorithm for solving a system of linear equations with a tridiagonal matrix, is constructed. The work of the constructed network is considered when solving a specific example. The number of operations carried out by the net in performing computations has been calculated and the performance of the parallel double sweep method implemented by the net is compared with the sequential double sweep method. We also consider representation of integer and real numbers by a SN and implementation of basic arithmetic operations over these numbers because the implementation of the algorithm requires floating point operations. GRAPHICAL ABSTRACT
{"title":"Realisation of the double sweep method by using a Sleptsov net","authors":"A. Kostikov, N. D. Zaitsev, O. Subotin","doi":"10.1080/17445760.2021.1945054","DOIUrl":"https://doi.org/10.1080/17445760.2021.1945054","url":null,"abstract":"The present article is focused on the application of Sleptsov net (SN) for solving system of linear algebraic equations with a tridiagonal matrix. SN, which implements the double sweep algorithm for solving a system of linear equations with a tridiagonal matrix, is constructed. The work of the constructed network is considered when solving a specific example. The number of operations carried out by the net in performing computations has been calculated and the performance of the parallel double sweep method implemented by the net is compared with the sequential double sweep method. We also consider representation of integer and real numbers by a SN and implementation of basic arithmetic operations over these numbers because the implementation of the algorithm requires floating point operations. GRAPHICAL ABSTRACT","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"516 - 534"},"PeriodicalIF":1.1,"publicationDate":"2021-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1945054","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"42797752","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}
Pub Date : 2021-06-27DOI: 10.1080/17445760.2021.1945055
T. Shmeleva, J. Owsinski, A. A. Lawan
Sleptsov nets are applied as a uniform language to specify models of unconventional computations and artificial intelligence systems. A technique for specification of neural networks, including multidimensional and multilayer networks of deep learning approach, using Sleptsov nets, is shown; the ways of specifying basic activation functions by Sleptsov net are discussed, the threshold and sigmoid functions implemented. A methodology of training neural networks is presented with the loss function minimisation, based on a run of a pair of interacting Sleptsov nets, the first net implementing the neural network based on data flow approach, while the second net solves the optimisation task by adjusting the weights of the first net by the gradient descend method. The optimising net uses the earlier developed technology of programming in Sleptsov nets with reverse control flow and the subnet call technique. Real numbers and arrays are represented as markings of a single place of a Sleptsov net. Hyperperformance is achieved because of the possibility of implementing mass parallel computations.
{"title":"Deep learning on Sleptsov nets","authors":"T. Shmeleva, J. Owsinski, A. A. Lawan","doi":"10.1080/17445760.2021.1945055","DOIUrl":"https://doi.org/10.1080/17445760.2021.1945055","url":null,"abstract":"Sleptsov nets are applied as a uniform language to specify models of unconventional computations and artificial intelligence systems. A technique for specification of neural networks, including multidimensional and multilayer networks of deep learning approach, using Sleptsov nets, is shown; the ways of specifying basic activation functions by Sleptsov net are discussed, the threshold and sigmoid functions implemented. A methodology of training neural networks is presented with the loss function minimisation, based on a run of a pair of interacting Sleptsov nets, the first net implementing the neural network based on data flow approach, while the second net solves the optimisation task by adjusting the weights of the first net by the gradient descend method. The optimising net uses the earlier developed technology of programming in Sleptsov nets with reverse control flow and the subnet call technique. Real numbers and arrays are represented as markings of a single place of a Sleptsov net. Hyperperformance is achieved because of the possibility of implementing mass parallel computations.","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"535 - 548"},"PeriodicalIF":1.1,"publicationDate":"2021-06-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1945055","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"42396734","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}
Pub Date : 2021-06-24DOI: 10.1080/17445760.2021.1941009
R. Ando, Y. Kadobayashi, H. Takakura
Science Information Network (SINET) is a Japanese academic backbone network for more than 800 research institutions and universities. In this paper, we present a multi-GPU-driven pipeline for handling huge session data of SINET. Our pipeline consists of ELK stack, multi-GPU server, and Splunk. A multi-GPU server is responsible for two procedures: discrimination and histogramming. Discrimination is dividing session data into ingoing/outgoing with subnet mask calculation and network address matching. Histogramming is grouping ingoing/outgoing session data into bins with map-reduce. In our architecture, we use GPU for the acceleration of ingress/egress discrimination of session data. Also, we use a tiling design pattern for building a two-stage map-reduce of CPU and GPU. Our multi-GPU-driven pipeline has succeeded in processing huge workloads of about 1.2–1.6 billion session streams (500–650 GB) within 24 hours. GRAPHICAL ABSTRACT
{"title":"Choice of parallelism: multi-GPU driven pipeline for huge academic backbone network","authors":"R. Ando, Y. Kadobayashi, H. Takakura","doi":"10.1080/17445760.2021.1941009","DOIUrl":"https://doi.org/10.1080/17445760.2021.1941009","url":null,"abstract":"Science Information Network (SINET) is a Japanese academic backbone network for more than 800 research institutions and universities. In this paper, we present a multi-GPU-driven pipeline for handling huge session data of SINET. Our pipeline consists of ELK stack, multi-GPU server, and Splunk. A multi-GPU server is responsible for two procedures: discrimination and histogramming. Discrimination is dividing session data into ingoing/outgoing with subnet mask calculation and network address matching. Histogramming is grouping ingoing/outgoing session data into bins with map-reduce. In our architecture, we use GPU for the acceleration of ingress/egress discrimination of session data. Also, we use a tiling design pattern for building a two-stage map-reduce of CPU and GPU. Our multi-GPU-driven pipeline has succeeded in processing huge workloads of about 1.2–1.6 billion session streams (500–650 GB) within 24 hours. GRAPHICAL ABSTRACT","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"609 - 622"},"PeriodicalIF":1.1,"publicationDate":"2021-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1941009","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45434205","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}
Pub Date : 2021-06-20DOI: 10.1080/17445760.2021.1939873
E. Anceaume, Yann Busnel, B. Sericola
We consider the problem of achieving uniform node sampling in large scale systems in presence of Byzantine nodes. This service offers a single simple primitive that returns, upon invocation, the identifier of a random node that belongs to the system. We first propose an omniscient strategy that processes on the fly an unbounded and arbitrarily biased input stream made of node identifiers exchanged within the system, and outputs a stream that preserves the uniformity property (same probability to appear in the sample). We show that this property holds despite any arbitrary bias introduced by the adversary. We then propose a strategy that is capable of approximating the omniscient strategy without requiring any prior knowledge on the composition of the input stream. We show through both theoretical analysis and extensive simulations that this strategy accurately approximates the omniscient one. We evaluate the resilience of the strategy by studying two representative attacks. We quantify the minimum number of identifiers that Byzantine nodes must insert in the input stream to prevent uniformity. Finally, we propose a new construction in series that allows to both increase the accuracy of a single sketch and decrease the time to converge to a uniform output stream.
{"title":"Byzantine-tolerant uniform node sampling service in large-scale networks","authors":"E. Anceaume, Yann Busnel, B. Sericola","doi":"10.1080/17445760.2021.1939873","DOIUrl":"https://doi.org/10.1080/17445760.2021.1939873","url":null,"abstract":"We consider the problem of achieving uniform node sampling in large scale systems in presence of Byzantine nodes. This service offers a single simple primitive that returns, upon invocation, the identifier of a random node that belongs to the system. We first propose an omniscient strategy that processes on the fly an unbounded and arbitrarily biased input stream made of node identifiers exchanged within the system, and outputs a stream that preserves the uniformity property (same probability to appear in the sample). We show that this property holds despite any arbitrary bias introduced by the adversary. We then propose a strategy that is capable of approximating the omniscient strategy without requiring any prior knowledge on the composition of the input stream. We show through both theoretical analysis and extensive simulations that this strategy accurately approximates the omniscient one. We evaluate the resilience of the strategy by studying two representative attacks. We quantify the minimum number of identifiers that Byzantine nodes must insert in the input stream to prevent uniformity. Finally, we propose a new construction in series that allows to both increase the accuracy of a single sketch and decrease the time to converge to a uniform output stream.","PeriodicalId":45411,"journal":{"name":"International Journal of Parallel Emergent and Distributed Systems","volume":"36 1","pages":"412 - 439"},"PeriodicalIF":1.1,"publicationDate":"2021-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1080/17445760.2021.1939873","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48862274","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：