1999
Journal article
Restricted
Parallelising the Mean Value Analysis algorithm
Gennaro C., King P. J. B.
The Mean Value Analysis (MVA) algorithm is one of the most popular for evaluating the performance of separable (or product form) queueing networks. Although its complexity is modest when jobs are indistinguishable, the introduction of different customer classes rapidly increases its computational cost. The problems of parallelising the algorithm while retaining its conceptual simplicity are examined. In particular, a parallel implementation of MVA on a distributed memory machine is developed using the MPI library for communication.Source: Transactions of the Society for Computer Simulation 16 (1999): 16–22. doi:10.1177/003754979907200304
DOI: 10.1177/003754979907200304
Metrics:
Gennaro C., King P. J. B.
The Mean Value Analysis (MVA) algorithm is one of the most popular for evaluating the performance of separable (or product form) queueing networks. Although its complexity is modest when jobs are indistinguishable, the introduction of different customer classes rapidly increases its computational cost. The problems of parallelising the algorithm while retaining its conceptual simplicity are examined. In particular, a parallel implementation of MVA on a distributed memory machine is developed using the MPI library for communication.Source: Transactions of the Society for Computer Simulation 16 (1999): 16–22. doi:10.1177/003754979907200304
DOI: 10.1177/003754979907200304
Metrics:
See at:
SIMULATION 
| sim.sagepub.com | CNR ExploRA
2002
Journal article
Restricted
Integrated Performance Models for SPMD Applications and MIMD Architectures
Cremonesi P., Gennaro C.
This paper introduces queuing network models for the performance analysis of SPMD applications executed on generalpurpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication, and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to anticipate the performance behavior of a parallel application as a function of the target architecture (i.e., number of processors, number of disks, I/O topology, etc).Source: IEEE transactions on parallel and distributed systems (Print) 13 (2002): 1320–1332. doi:10.1109/TPDS.2002.1158268
DOI: 10.1109/tpds.2002.1158268
DOI: 10.1109/tpds.2002.1019862
Metrics:
Cremonesi P., Gennaro C.
This paper introduces queuing network models for the performance analysis of SPMD applications executed on generalpurpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication, and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to anticipate the performance behavior of a parallel application as a function of the target architecture (i.e., number of processors, number of disks, I/O topology, etc).Source: IEEE transactions on parallel and distributed systems (Print) 13 (2002): 1320–1332. doi:10.1109/TPDS.2002.1158268
DOI: 10.1109/tpds.2002.1158268
DOI: 10.1109/tpds.2002.1019862
Metrics:
See at:
IEEE Transactions on Parallel and Distributed Systems | IEEE Transactions on Parallel and Distributed Systems | ieeexplore.ieee.org | CNR ExploRA
2008
Journal article
Restricted
Regia: a metadata editor for audiovisual documents
Gennaro C.
Although the Metadata Editor is an important part of any digital library, it becomes fundamental in the presence of audiovisual content. This is because the metadata produced by automated support tools (such as speech recognizers and shot detection procedures) is error-prone and often needs correction. In addition, scenes are manually annotated. This paper describes Regia, a prototype application for manually editing metadata for audiovisual documents developed in the ECHO project. Regia allows the user to manually edit textual metadata and to hierarchically organize the segmentation of the audiovisual content. An important feature of this metadata editor is that it is not hard-wired with a particular metadata attributes set. To achieve this feature the XML schema of the metadata model is used by the editor as a configuration file.Source: Multimedia tools and applications 36 (2008): 185–201. doi:10.1007/s11042-007-0129-4
DOI: 10.1007/s11042-007-0129-4
Metrics:
Gennaro C.
Although the Metadata Editor is an important part of any digital library, it becomes fundamental in the presence of audiovisual content. This is because the metadata produced by automated support tools (such as speech recognizers and shot detection procedures) is error-prone and often needs correction. In addition, scenes are manually annotated. This paper describes Regia, a prototype application for manually editing metadata for audiovisual documents developed in the ECHO project. Regia allows the user to manually edit textual metadata and to hierarchically organize the segmentation of the audiovisual content. An important feature of this metadata editor is that it is not hard-wired with a particular metadata attributes set. To achieve this feature the XML schema of the metadata model is used by the editor as a configuration file.Source: Multimedia tools and applications 36 (2008): 185–201. doi:10.1007/s11042-007-0129-4
DOI: 10.1007/s11042-007-0129-4
Metrics:
See at:
Multimedia Tools and Applications | www.springerlink.com | CNR ExploRA
2004
Journal article
Open Access
GHT*: a Peer-to-Peer System for Metric Data
Batko M., Gennaro C., Zezula P.
GHT* is a scalable and distributed similarity search structure that has been specifically designed to support metric space objects. Our structure is based on the P2P communication paradigm and it is scalable in that it distributes the data over more and more independent peer computers. By exploiting parallelism in a dynamic network of computers, the query execution scales up very well considering both the number of distance computations and the hop count between the peers. Updates are performed locally and a node splitting never requires sending multiple messages to many peers.Source: ERCIM news 59 (2004): 65–66.
Batko M., Gennaro C., Zezula P.
GHT* is a scalable and distributed similarity search structure that has been specifically designed to support metric space objects. Our structure is based on the P2P communication paradigm and it is scalable in that it distributes the data over more and more independent peer computers. By exploiting parallelism in a dynamic network of computers, the query execution scales up very well considering both the number of distance computations and the hop count between the peers. Updates are performed locally and a node splitting never requires sending multiple messages to many peers.Source: ERCIM news 59 (2004): 65–66.
See at:
www.ercim.eu 
| CNR ExploRA
2002
Conference article
Restricted
A Metric Index for Approximate Text Management
Dohnal V., Gennaro C., Zezula P.
Text collections of data need not only search support for identical objects, but the approximate matching is even more important. A suitable metric to such a task is the edit distance measure. However, the quadratic computational complexity of edit distance prevents from applying naive storage organizations, such as the sequential search, and more sophisticated search structures must be applied. We have investigated the properties of the D-index to approximate searching and matching in text databases. The experiments confirm a very good performance for retrieving close objects and sub-linear scalability to process large files. Even the similarity joins can be performed efficiently.Source: International Conference Information Systems and Databases (ISDB 2002), pp. 37–42, Tokyo, Japan, 5-27 September 2002
Dohnal V., Gennaro C., Zezula P.
Text collections of data need not only search support for identical objects, but the approximate matching is even more important. A suitable metric to such a task is the edit distance measure. However, the quadratic computational complexity of edit distance prevents from applying naive storage organizations, such as the sequential search, and more sophisticated search structures must be applied. We have investigated the properties of the D-index to approximate searching and matching in text databases. The experiments confirm a very good performance for retrieving close objects and sub-linear scalability to process large files. Even the similarity joins can be performed efficiently.Source: International Conference Information Systems and Databases (ISDB 2002), pp. 37–42, Tokyo, Japan, 5-27 September 2002
See at:
dblp.uni-trier.de | CNR ExploRA
2008
Conference article
Restricted
A content-addressable network for similarity join in metric spaces
Gennaro C.
Similarity join is an interesting complement of the well-established similarity range and nearest neighbors search primitives in metric spaces. However, the quadratic computational complexity of similarity join prevents from applications on large data collections. We present MCAN+, an extension of MCAN (a Content-Addressable Network for metric objects) to support similarity self join queries. The challenge of the proposed approach is to address the problem of the intrinsic quadratic complexity of similarity joins, with the aim of limiting the elaboration time, by involving an increasing number of computational nodes as the dataset size grows. To test the scalability of MCAN+, we used a real-life dataset of color features extracted from one million images of the Flickr photo sharing website.Source: Third Interational ICST Conference on Scalable Information Systems. Infoscale'08, Vico Equense, Italy, 4-6 June 2008
Gennaro C.
Similarity join is an interesting complement of the well-established similarity range and nearest neighbors search primitives in metric spaces. However, the quadratic computational complexity of similarity join prevents from applications on large data collections. We present MCAN+, an extension of MCAN (a Content-Addressable Network for metric objects) to support similarity self join queries. The challenge of the proposed approach is to address the problem of the intrinsic quadratic complexity of similarity joins, with the aim of limiting the elaboration time, by involving an increasing number of computational nodes as the dataset size grows. To test the scalability of MCAN+, we used a real-life dataset of color features extracted from one million images of the Flickr photo sharing website.Source: Third Interational ICST Conference on Scalable Information Systems. Infoscale'08, Vico Equense, Italy, 4-6 June 2008
See at:
dl.acm.org | CNR ExploRA
2008
Conference article
Unknown
A methodology for building and querying an ontology representing data and multimedia sources
Beneventano D., Gennaro C., Guerra F.
Managing data and multimedia sources with a unique tool is a challenging issue. In this paper, the capabilities of the MOMIS integration system and the MILOS multimedia content management system are coupled, thus providing a methodology and a tool for building and querying a populated ontology representing data and multimedia sources.Source: 4th International VLDB Workshop on Ontology-based Techniques for DataBases in Information Systems and Knowledge Systems, pp. 37–40, New Zealand, 23 August 2008
Beneventano D., Gennaro C., Guerra F.
Managing data and multimedia sources with a unique tool is a challenging issue. In this paper, the capabilities of the MOMIS integration system and the MILOS multimedia content management system are coupled, thus providing a methodology and a tool for building and querying a populated ontology representing data and multimedia sources.Source: 4th International VLDB Workshop on Ontology-based Techniques for DataBases in Information Systems and Knowledge Systems, pp. 37–40, New Zealand, 23 August 2008
See at: CNR ExploRA
2003
Contribution to conference
Open Access
Scalable and Distributed Similarity Search in Metric Spaces
Batko M., Gennaro C., Zezula P.
In this paper we propose a new access structure, called GHT*, based on generalized hyperplane tree (GHT) and distributed dynamic hashing (DDH) techniques. GHT* is a distributed structure which allows to perform range search in a metric space according to a distance function d. The structure does not require a central directory and it is able to gracefully scale through splits of one bucket at a time.Source: Workshop on Distributed Data and Structures, n.5, pp. 107–116, Thessaloniki, Greece, 13-14 June 2003
Batko M., Gennaro C., Zezula P.
In this paper we propose a new access structure, called GHT*, based on generalized hyperplane tree (GHT) and distributed dynamic hashing (DDH) techniques. GHT* is a distributed structure which allows to perform range search in a metric space according to a distance function d. The structure does not require a central directory and it is able to gracefully scale through splits of one bucket at a time.Source: Workshop on Distributed Data and Structures, n.5, pp. 107–116, Thessaloniki, Greece, 13-14 June 2003
See at:
ISTI Repository | CNR ExploRA
2004
Contribution to conference
Open Access
A Scalable Nearest Neighbor Search in P2P Systems
Batko M., Gennaro C., Zezula P.
Similarity search in metric spaces represents an important paradigm for content-based retrieval of many applications. Existing centralized search structures can speed-up retrieval, but they do not scale up to large volume of data because the response time is linearly increasing with the size of the searched file. In this article, we study the problem of executing the nearest neighbor(s) queries in a distributed metric structure, which is based on the P2P communication paradigm and the generalized hyperplane partitioning. By exploiting parallelism in a dynamic network of computers, the query execution scales up very well considering both the number of distance computations and the hop count between the peers. Results are verified by experiments on real-life data sets.Source: International Workshop on Databases, Information Systems and Peer-to-Peer Computing, pp. 1–14, Toronto, Canada, 29-30 August 2004
Batko M., Gennaro C., Zezula P.
Similarity search in metric spaces represents an important paradigm for content-based retrieval of many applications. Existing centralized search structures can speed-up retrieval, but they do not scale up to large volume of data because the response time is linearly increasing with the size of the searched file. In this article, we study the problem of executing the nearest neighbor(s) queries in a distributed metric structure, which is based on the P2P communication paradigm and the generalized hyperplane partitioning. By exploiting parallelism in a dynamic network of computers, the query execution scales up very well considering both the number of distance computations and the hop count between the peers. Results are verified by experiments on real-life data sets.Source: International Workshop on Databases, Information Systems and Peer-to-Peer Computing, pp. 1–14, Toronto, Canada, 29-30 August 2004
See at:
ISTI Repository | CNR ExploRA
2003
Software
Unknown
Regia: a metadata editor for Audio/Video Document developed within the EU ECHO project
Gennaro C.
Regia is a metadata editor for Audio/Video Document developed within the EU ECHO project. Regia generally works connected to the ECHO Digital Library System, which unfortunately is not available anymore. However, the program is also able to work in standalone modality. The metadata model of ECHO is based on the IFLA model, a general conceptual framework used to describe heterogeneous digital media resources. This metadata model is composed of four levels describing different aspects of intellectual or artistic endeavour: AVDocument, Expression, Media, and Storage. The entities of the model are organized in a structure that reflects the hierarchical order of the entities from the top level (work) to the bottom (item). Figure 1 shows a schematic representation of the ECHO Metadata Model. As it is possible to see, the metadata which belong to different classes comprised in the model, are logically divided in two sets Bibliographic Metadata and Time/Space related Metadata. This classification is also reflected in the Metadata Editor interface.
Gennaro C.
Regia is a metadata editor for Audio/Video Document developed within the EU ECHO project. Regia generally works connected to the ECHO Digital Library System, which unfortunately is not available anymore. However, the program is also able to work in standalone modality. The metadata model of ECHO is based on the IFLA model, a general conceptual framework used to describe heterogeneous digital media resources. This metadata model is composed of four levels describing different aspects of intellectual or artistic endeavour: AVDocument, Expression, Media, and Storage. The entities of the model are organized in a structure that reflects the hierarchical order of the entities from the top level (work) to the bottom (item). Figure 1 shows a schematic representation of the ECHO Metadata Model. As it is possible to see, the metadata which belong to different classes comprised in the model, are logically divided in two sets Bibliographic Metadata and Time/Space related Metadata. This classification is also reflected in the Metadata Editor interface.
See at: CNR ExploRA
2009
Conference article
Unknown
Scalable similarity self join in a metric DHT system
Gennaro C.
Efficient processing of similarity joins is important for a large class of data analysis and data-mining applications. This primitive finds all pairs of records within a predefined distance threshold of each other. We present MCAN+, an extension of MCAN (a Content-Addressable Network for metric objects) to support similarity self join queries. The challenge of the proposed approach is to address the problem of the intrinsic quadratic complexity of similarity joins, with the aim of bounding the elaboration time, by involving an increasing number of computational nodes as the dataset size grows. To test the scalability of MCAN+, we used a real-life dataset of color features extracted from one million images of the Flickr photo sharing website.Source: 17th Italian Symposium on Advanced Database Systems, pp. 81–88, Camogli, Genova, 21-24 June 2009
Gennaro C.
Efficient processing of similarity joins is important for a large class of data analysis and data-mining applications. This primitive finds all pairs of records within a predefined distance threshold of each other. We present MCAN+, an extension of MCAN (a Content-Addressable Network for metric objects) to support similarity self join queries. The challenge of the proposed approach is to address the problem of the intrinsic quadratic complexity of similarity joins, with the aim of bounding the elaboration time, by involving an increasing number of computational nodes as the dataset size grows. To test the scalability of MCAN+, we used a real-life dataset of color features extracted from one million images of the Flickr photo sharing website.Source: 17th Italian Symposium on Advanced Database Systems, pp. 81–88, Camogli, Genova, 21-24 June 2009
See at: CNR ExploRA
2005
Report
Open Access
Regia: a metadata editor for audiovisual documents
Gennaro C.
Although the Metadata Editor is an important part of any digital library, it becomes fundamental in the presence of audiovisual content. This is because the metadata produced by automated support tools (such as speech recognizers and shot detection procedures) is error-prone and often needs correction. In addition, scenes are manually annotated. This paper describes Regia, a prototype application for manual editing metadata for audiovisual documents developed in the ECHO project. Regia allows the user to manually edit textual metadata and to hierarchically organize the segmentation of the audiovisual content. An important feature of this metadata editor is that it is not hard--wired with a particular metadata attributes set; for this purpose the XML schema of the metadata model is used by the editor as configuration file.Source: ISTI Technical reports, 2005
Gennaro C.
Although the Metadata Editor is an important part of any digital library, it becomes fundamental in the presence of audiovisual content. This is because the metadata produced by automated support tools (such as speech recognizers and shot detection procedures) is error-prone and often needs correction. In addition, scenes are manually annotated. This paper describes Regia, a prototype application for manual editing metadata for audiovisual documents developed in the ECHO project. Regia allows the user to manually edit textual metadata and to hierarchically organize the segmentation of the audiovisual content. An important feature of this metadata editor is that it is not hard--wired with a particular metadata attributes set; for this purpose the XML schema of the metadata model is used by the editor as configuration file.Source: ISTI Technical reports, 2005
See at:
ISTI Repository | CNR ExploRA
2001
Report
Unknown
Integrated performance models for SPMD applications and MIMD architectures
Cremonesi P., Gennaro C.
The report introduces queuing network models for the performance analysis of SPMD applications executed on general-purpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to predict the performance of parallel applications in early stages of software development.Source: ISTI Technical reports, pp.1–19, 2001
Cremonesi P., Gennaro C.
The report introduces queuing network models for the performance analysis of SPMD applications executed on general-purpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to predict the performance of parallel applications in early stages of software development.Source: ISTI Technical reports, pp.1–19, 2001
See at: CNR ExploRA
2009
Report
Open Access
A theoretical approach to the self similarity join in a distributed enviroment
Gennaro C.
Efficient processing of similarity joins is important for a large class of data analysis and data-mining applications. This primitive finds all pairs of records within a predefined distance threshold of each other. However, most of the existing approaches have been based on spatial join techniques designed primarily for data in a vector space. Treating data collections as metric objects brings a great advantage in generality, because a single metric technique can be applied to many specific search problems quite different in nature. In this paper, we concentrate our attention on a special form of join, the Self Similarity Join, which retrieves pairs from the same dataset. In particular, we consider the case in which the dataset is split into subsets that are searched for self similarity join independently (e.g, in a distributed computing environment). To this end, we formalize the abstract concept of epsilon-Cover, prove its correctness, and demonstrate its effectiveness by applying it to two real implementations on a real-life large dataset.Source: ISTI Technical reports, 2009
Gennaro C.
Efficient processing of similarity joins is important for a large class of data analysis and data-mining applications. This primitive finds all pairs of records within a predefined distance threshold of each other. However, most of the existing approaches have been based on spatial join techniques designed primarily for data in a vector space. Treating data collections as metric objects brings a great advantage in generality, because a single metric technique can be applied to many specific search problems quite different in nature. In this paper, we concentrate our attention on a special form of join, the Self Similarity Join, which retrieves pairs from the same dataset. In particular, we consider the case in which the dataset is split into subsets that are searched for self similarity join independently (e.g, in a distributed computing environment). To this end, we formalize the abstract concept of epsilon-Cover, prove its correctness, and demonstrate its effectiveness by applying it to two real implementations on a real-life large dataset.Source: ISTI Technical reports, 2009
See at:
ISTI Repository | CNR ExploRA
2003
Journal article
Restricted
Effective Scheduling of Detached Rules in Active Databases
S. Ceri, C. Gennaro, S. Paraboschi, G. Serazzi
While triggers have become a classical ingredient of relational database systems, research in active databases is aiming at extending the functionality and expressive power of active rules beyond the scope of relational triggers. One of the most important current trend concerns the support of detached active rules, i.e., of rules which are executed as separate transactions, running outside of the scope of the transaction which generates the triggering event. Detached rules have important applications in workflow management and global integrity maintenance across transactions. One of the main issues in designing the rule engine for detached rules is determining their optimal scheduling. In this paper, we study the performance of a detached rule scheduler whose objective is to minimize the interference of detached rule execution with regard to the normal transactional load. This objective is achieved by executing detached rules at given periods of time and by assigning them a fixed amount of dedicated threads; we study the performance of the scheduler relative to the two most critical design parameters, the frequency of execution of the scheduler, and the number of dedicated execution threads.Source: IEEE transactions on knowledge and data engineering (Print) 15 (2003): 2–13. doi:10.1109/TKDE.2003.1161578
DOI: 10.1109/tkde.2003.1161578
Metrics:
S. Ceri, C. Gennaro, S. Paraboschi, G. Serazzi
While triggers have become a classical ingredient of relational database systems, research in active databases is aiming at extending the functionality and expressive power of active rules beyond the scope of relational triggers. One of the most important current trend concerns the support of detached active rules, i.e., of rules which are executed as separate transactions, running outside of the scope of the transaction which generates the triggering event. Detached rules have important applications in workflow management and global integrity maintenance across transactions. One of the main issues in designing the rule engine for detached rules is determining their optimal scheduling. In this paper, we study the performance of a detached rule scheduler whose objective is to minimize the interference of detached rule execution with regard to the normal transactional load. This objective is achieved by executing detached rules at given periods of time and by assigning them a fixed amount of dedicated threads; we study the performance of the scheduler relative to the two most critical design parameters, the frequency of execution of the scheduler, and the number of dedicated execution threads.Source: IEEE transactions on knowledge and data engineering (Print) 15 (2003): 2–13. doi:10.1109/TKDE.2003.1161578
DOI: 10.1109/tkde.2003.1161578
Metrics:
See at:
IEEE Transactions on Knowledge and Data Engineering | ieeexplore.ieee.org | CNR ExploRA
1998
Conference article
Restricted
I/O performance in hybrid MIMD+SIMD machines
Cremonesi P., Gennaro C., Marega R.
The Rewrite Rule Machine (RRM) is a massively parallel MIMD/SIMD computer designed with the explicit purpose of supporting very-high-level parallel programming with rewrite rules. The RRM's node architecture consists of a SIMD processor, a SIMD controller, local memory, and network and I/O interfaces. A 64-node cluster board is already an attractive RRM system capable of extremely high performance on a variety of applications. A cluster is SIMD at the node level, but it is MIMD at the system level to flexibly exploit the parallelism of complex nonhomogeneous applications. In addition to reporting detailed simulation experiments used to validate the node design, we measure the performance of an RRM cluster on three relevant applications.Source: International Conference and Exhibition (HPCN Europe 1998), pp. 688–697, Amsterdam, 1998
DOI: 10.1007/bfb0037196
Metrics:
Cremonesi P., Gennaro C., Marega R.
The Rewrite Rule Machine (RRM) is a massively parallel MIMD/SIMD computer designed with the explicit purpose of supporting very-high-level parallel programming with rewrite rules. The RRM's node architecture consists of a SIMD processor, a SIMD controller, local memory, and network and I/O interfaces. A 64-node cluster board is already an attractive RRM system capable of extremely high performance on a variety of applications. A cluster is SIMD at the node level, but it is MIMD at the system level to flexibly exploit the parallelism of complex nonhomogeneous applications. In addition to reporting detailed simulation experiments used to validate the node design, we measure the performance of an RRM cluster on three relevant applications.Source: International Conference and Exhibition (HPCN Europe 1998), pp. 688–697, Amsterdam, 1998
DOI: 10.1007/bfb0037196
Metrics:
See at:
doi.org | www.scopus.com | CNR ExploRA
1999
Conference article
Restricted
Performance models for I/O bound SPMD applications on clusters of workstations
Gennaro C.
Clusters of workstations represent today a satisfactory alternative to MPPs and supercomputers in many areas of application. The rapidly reduction of the cost of high performance workstations/PCs makes this technology ever more available. Moreover, new concepts for the integration of individual workstations through Local Area Networks are emerging. High speed interconnection networks and optimized protocol system architectures are the most important objectives of current research in this field of study. In this contribution, we attempt to propose a simple but effective performance model of systems with distributed computational and I/O resources when executing parallel scientific applications characterized by communication bursts and by intensive I/O phases. By means of queueing network techniques, the analysis of the model leads to the definition of a speedup surface which captures the relative influence of processors and disks parallelism in the performance of applications that alternates computations and I/O operations in a cyclic fashion.Source: Seventh Euromicro Workshop on Parallel and Distributed Processing (PDP '99), pp. 263–270, University of Madeira, Funchal, Portugal, 3/2/1999
DOI: 10.1109/empdp.1999.746686
Metrics:
Gennaro C.
Clusters of workstations represent today a satisfactory alternative to MPPs and supercomputers in many areas of application. The rapidly reduction of the cost of high performance workstations/PCs makes this technology ever more available. Moreover, new concepts for the integration of individual workstations through Local Area Networks are emerging. High speed interconnection networks and optimized protocol system architectures are the most important objectives of current research in this field of study. In this contribution, we attempt to propose a simple but effective performance model of systems with distributed computational and I/O resources when executing parallel scientific applications characterized by communication bursts and by intensive I/O phases. By means of queueing network techniques, the analysis of the model leads to the definition of a speedup surface which captures the relative influence of processors and disks parallelism in the performance of applications that alternates computations and I/O operations in a cyclic fashion.Source: Seventh Euromicro Workshop on Parallel and Distributed Processing (PDP '99), pp. 263–270, University of Madeira, Funchal, Portugal, 3/2/1999
DOI: 10.1109/empdp.1999.746686
Metrics:
See at:
doi.org | www.computer.org | CNR ExploRA
2002
Journal article
Restricted
Integrated Performance Models for SPMD Applications and MIMD Architectures
Cremonesi P., Gennaro C.
This paper introduces queuing network models for the performance analysis of SPMD applications executed on generalpurpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication, and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to anticipate the performance behavior of a parallel application as a function of the target architecture (i.e., number of processors, number of disks, I/O topology, etc).Source: IEEE transactions on parallel and distributed systems (Print) 13 (2002): 745–757. doi:10.1109/TPDS.2002.1019862
DOI: 10.1109/tpds.2002.1019862
DOI: 10.1109/tpds.2002.1158268
Metrics:
Cremonesi P., Gennaro C.
This paper introduces queuing network models for the performance analysis of SPMD applications executed on generalpurpose parallel architectures such as MIMD and clusters of workstations. The models are based on the pattern of computation, communication, and I/O operations of typical parallel applications. Analysis of the models leads to the definition of speedup surfaces which capture the relative influence of processors and I/O parallelism and show the effects of different hardware and software components on the performance. Since the parameters of the models correspond to measurable program and hardware characteristics, the models can be used to anticipate the performance behavior of a parallel application as a function of the target architecture (i.e., number of processors, number of disks, I/O topology, etc).Source: IEEE transactions on parallel and distributed systems (Print) 13 (2002): 745–757. doi:10.1109/TPDS.2002.1019862
DOI: 10.1109/tpds.2002.1019862
DOI: 10.1109/tpds.2002.1158268
Metrics:
See at:
IEEE Transactions on Parallel and Distributed Systems | IEEE Transactions on Parallel and Distributed Systems | ieeexplore.ieee.org | CNR ExploRA
1999
Doctoral thesis
Unknown
Integrated models for SIMD, MIMD and hybrid parallel architectures
Gennaro C.
The purpose of massively parallel machines is to solve problems in various scientific/engineering domains that would have unacceptable processing times if solved with a traditional single processor machine. The performance analysis has emerged as a key area of research in the field of parallel architectures, ever since the first high performance machines appeared on the market. This appeared ever more true as the trend of the high performance computer moved from using vector and array processors to using multiprocessor architectures. Many factors influence the performance of parallel applications. Hardware or software components, or both can be involved. Examples are: processor speed, processor communication network, I/O architecture, sequential code, communication network contention and synchronization protocols, data partitioning, etc. The first effort to model the performance of parallel programs is due to Amdahl and the well known homonymous law that relates the number of processors to the bound of the speedup which may be expected by parallel processing~cite{AM91}. This bound has proved useful in shaping our understanding of parallel system because it strikes a useful balance between simplicity and precision. Several amendments and extensions to Amdahl's law have been proposed, each appropriate for different purposes. Gustafson et al.~cite{GU88ACM,GU88,GMB88} introduce the concept of scaled speedup as a measure of how well an algorithm scales when the number of processors is multiplied by $k$. Flat et al.~cite{FK89} investigate the impact of synchronization and communication overhead on the performance of parallel processors. Eager et al.~cite{ELZ89} studied speedup versus efficiency. Wu et al.~cite{WL98} propose a formal definition of scalability and discuss scalability of cluster systems. Finally, Rosti et al.~cite{RSSS98} extends Amdahl's law to three--dimensional space to include processors and disks. However, these studies are limited by the fact that the speedup formulations do not include any explicit parameters that reference the communication and I/O overhead or the type of hardware used for executing the parallel application. Other approaches use task graph models in order to represent the intertask precedence relations within a program, the task execution times, and the times necessary to transfer data or other information from one task to other during execution. In the case of program control structures that can be represented by ``series--parallel'' task graphs, such as the fork--join structure, methodology to predict the best speedup which can be obtained with such program has been developed. In particular, when the task execution times are deterministic, Coffman et al.~cite{CD73} used critical path analysis to find the program completion time. If the task execution times are stochastic and the number of processors is infinite, the probability distribution time can be determined by a straightforward but in general very costly computation~cite{HH79,BA74}. General acyclic random graph models are presented in~cite{FKM83,MN84,GNPT86}. Gelenbe~cite{Gelenbe89} generalizes the task graph models so that it is possible to take into account the effect of communication times between the tasks. For more realistic cases where the number of processors is smaller than the number of tasks in a program, a queueing network approach can be used. Approximate models are presented in~cite{CL87,GL88,GMSW86} and an exact model is proposed by Baccelli et al.~cite{BZ90}. Task graphs, as general acyclic graphs, allow the description of sequential or parallel execution, synchronization, and spawning of task. Nevertheless, the multiprocessor systems considered by this kind of models have a generic architecture with a finite number of processors sharing a central memory. This work proposes a new model for parallel systems with distributed computational and I/O resources when executing parallel applications characterized by cyclic computation bursts and intensive I/O bursts. By means of queuing network techniques, the analysis of the model leads to the definition of a generic model that allows simulation of the behaviour of several parallel architectures. Traditionally, speedup is defined as the ratio of the elapsed time when executing a program on a single processor to the execution time when $p$ processors are available. We extend this definition to include the number $d$ of {it I/O nodes}. Because parallel program execution can be partitioned into two distinct phases, denoted as computation burst and I/O burst, we consider the $p$ processors devoted to the execution of the computation burst and the $d$ I/O nodes, to the corresponding I/O burst. An I/O node can represent a disk or a SIMD processor according to the parallel machine modeled. For instance consider a hybrid architecture in which a moderate number of processors (10-20) are arranged in a MIMD way and all of them are ``boosted'' by massively parallel SIMD arrays, used for the the execution of tasks that are particularly suitable for SIMD architectures. The clusters of SIMD processors is connected to the MIMD node through a high speed I/O channel. In this scenario, several questions may arise, such as: how does the program speedup scale with the number of SIMD processors? How much does the I/O channel impact to performance of the program? Is there any advantage in using the SIMD processors or it is better to use only the MIMD processors to perform the SIMD tasks? The approach we propose in this thesis is, to the best author's knowledge, the first attempt to address these kinds of issues. The purpose of this thesis is to develop analytical performance models that captures the behavior of numerous applications running on a different parallel architectures. The idea of our approach is to use queueing network models. This type of model has the advantage of providing a ``white--box'' view of the system to study. Because the parameters of our model correspond to measurable program characteristics, we can use the model in order to estimate the execution times for different number of processors and/or I/O nodes. nition of a generic model that allows simulation of the behaviour of several parallel architectures.
Gennaro C.
The purpose of massively parallel machines is to solve problems in various scientific/engineering domains that would have unacceptable processing times if solved with a traditional single processor machine. The performance analysis has emerged as a key area of research in the field of parallel architectures, ever since the first high performance machines appeared on the market. This appeared ever more true as the trend of the high performance computer moved from using vector and array processors to using multiprocessor architectures. Many factors influence the performance of parallel applications. Hardware or software components, or both can be involved. Examples are: processor speed, processor communication network, I/O architecture, sequential code, communication network contention and synchronization protocols, data partitioning, etc. The first effort to model the performance of parallel programs is due to Amdahl and the well known homonymous law that relates the number of processors to the bound of the speedup which may be expected by parallel processing~cite{AM91}. This bound has proved useful in shaping our understanding of parallel system because it strikes a useful balance between simplicity and precision. Several amendments and extensions to Amdahl's law have been proposed, each appropriate for different purposes. Gustafson et al.~cite{GU88ACM,GU88,GMB88} introduce the concept of scaled speedup as a measure of how well an algorithm scales when the number of processors is multiplied by $k$. Flat et al.~cite{FK89} investigate the impact of synchronization and communication overhead on the performance of parallel processors. Eager et al.~cite{ELZ89} studied speedup versus efficiency. Wu et al.~cite{WL98} propose a formal definition of scalability and discuss scalability of cluster systems. Finally, Rosti et al.~cite{RSSS98} extends Amdahl's law to three--dimensional space to include processors and disks. However, these studies are limited by the fact that the speedup formulations do not include any explicit parameters that reference the communication and I/O overhead or the type of hardware used for executing the parallel application. Other approaches use task graph models in order to represent the intertask precedence relations within a program, the task execution times, and the times necessary to transfer data or other information from one task to other during execution. In the case of program control structures that can be represented by ``series--parallel'' task graphs, such as the fork--join structure, methodology to predict the best speedup which can be obtained with such program has been developed. In particular, when the task execution times are deterministic, Coffman et al.~cite{CD73} used critical path analysis to find the program completion time. If the task execution times are stochastic and the number of processors is infinite, the probability distribution time can be determined by a straightforward but in general very costly computation~cite{HH79,BA74}. General acyclic random graph models are presented in~cite{FKM83,MN84,GNPT86}. Gelenbe~cite{Gelenbe89} generalizes the task graph models so that it is possible to take into account the effect of communication times between the tasks. For more realistic cases where the number of processors is smaller than the number of tasks in a program, a queueing network approach can be used. Approximate models are presented in~cite{CL87,GL88,GMSW86} and an exact model is proposed by Baccelli et al.~cite{BZ90}. Task graphs, as general acyclic graphs, allow the description of sequential or parallel execution, synchronization, and spawning of task. Nevertheless, the multiprocessor systems considered by this kind of models have a generic architecture with a finite number of processors sharing a central memory. This work proposes a new model for parallel systems with distributed computational and I/O resources when executing parallel applications characterized by cyclic computation bursts and intensive I/O bursts. By means of queuing network techniques, the analysis of the model leads to the definition of a generic model that allows simulation of the behaviour of several parallel architectures. Traditionally, speedup is defined as the ratio of the elapsed time when executing a program on a single processor to the execution time when $p$ processors are available. We extend this definition to include the number $d$ of {it I/O nodes}. Because parallel program execution can be partitioned into two distinct phases, denoted as computation burst and I/O burst, we consider the $p$ processors devoted to the execution of the computation burst and the $d$ I/O nodes, to the corresponding I/O burst. An I/O node can represent a disk or a SIMD processor according to the parallel machine modeled. For instance consider a hybrid architecture in which a moderate number of processors (10-20) are arranged in a MIMD way and all of them are ``boosted'' by massively parallel SIMD arrays, used for the the execution of tasks that are particularly suitable for SIMD architectures. The clusters of SIMD processors is connected to the MIMD node through a high speed I/O channel. In this scenario, several questions may arise, such as: how does the program speedup scale with the number of SIMD processors? How much does the I/O channel impact to performance of the program? Is there any advantage in using the SIMD processors or it is better to use only the MIMD processors to perform the SIMD tasks? The approach we propose in this thesis is, to the best author's knowledge, the first attempt to address these kinds of issues. The purpose of this thesis is to develop analytical performance models that captures the behavior of numerous applications running on a different parallel architectures. The idea of our approach is to use queueing network models. This type of model has the advantage of providing a ``white--box'' view of the system to study. Because the parameters of our model correspond to measurable program characteristics, we can use the model in order to estimate the execution times for different number of processors and/or I/O nodes. nition of a generic model that allows simulation of the behaviour of several parallel architectures.
See at: CNR ExploRA
2003
Journal article
Restricted
D-Index: Distance Searching Index for Metric Data Sets, Multimedia Tools and Applications
Dohnal V., Gennaro C., Savino P., Zezula P.
In order to speedup retrieval in large collections of data, index structures partition the data into subsets so that query requests can be evaluated without examining the entire collection. As the complexity of modern data types grows, metric spaces have become a popular paradigm for similarity retrieval. We propose a new index structure, called D-Index, that combines a novel clustering technique and the pivot-based distance searching strategy to speed up execution of similarity range and nearest neighbor queries for large files with objects stored in disk memories. We have qualitatively analyzed D-Index and verified its properties on actual implementation. We have also compared D-Index with other index structures and demonstrated its superiority on several real-life data sets. Contrary to tree organizations, the D-Index structure is suitable for dynamic environments with a high rate of delete/insert operations.Source: Multimedia tools and applications 21 (2003): 9–33. doi:10.1023/A:1025026030880
DOI: 10.1023/a:1025026030880
DOI: 10.1023/a%3a1025026030880
Metrics:
Dohnal V., Gennaro C., Savino P., Zezula P.
In order to speedup retrieval in large collections of data, index structures partition the data into subsets so that query requests can be evaluated without examining the entire collection. As the complexity of modern data types grows, metric spaces have become a popular paradigm for similarity retrieval. We propose a new index structure, called D-Index, that combines a novel clustering technique and the pivot-based distance searching strategy to speed up execution of similarity range and nearest neighbor queries for large files with objects stored in disk memories. We have qualitatively analyzed D-Index and verified its properties on actual implementation. We have also compared D-Index with other index structures and demonstrated its superiority on several real-life data sets. Contrary to tree organizations, the D-Index structure is suitable for dynamic environments with a high rate of delete/insert operations.Source: Multimedia tools and applications 21 (2003): 9–33. doi:10.1023/A:1025026030880
DOI: 10.1023/a:1025026030880
DOI: 10.1023/a%3a1025026030880
Metrics:
See at:
Multimedia Tools and Applications | CNR ExploRA