Olav Lysne - Academia.edu (original) (raw)

Papers by Olav Lysne

In this paper we present a parameterized SCI node model implemented in OPNET Modeler. It contains... more In this paper we present a parameterized SCI node model implemented in OPNET Modeler. It contains a full implementation of the go-bit based flow control, as well as the A-B aging scheme of the retry protocol. The simulation environment offers a highly graphical user interface, allowing e.g. the formation of ringlets and larger topologies through drag and deop functionality. A standard distribution of OPNET contains models of most standard communication technologies like Ethernet, FDDI, ATM, etc. This facilitates easy integration of our node with models of other technologies, enabling simulation of heterogeneous systems with very limited additional development.

The ability to tolerate one failing link in communication networks is sufficient for many practic... more The ability to tolerate one failing link in communication networks is sufficient for many practical purposes. One-fault tolerance can also be achieved at much lower cost than methods that can guarantee tolerance of multiple faults. We consider wormhole routed meshes with two different routing algorithms - dimension order and positive first - and for each of these we propose two

Proceedings of the 17th International Symposium on Parallel and Distributed Processing, 2003

In recent years, we have seen an increased interest in irregular network topologies. This has bee... more In recent years, we have seen an increased interest in irregular network topologies. This has been spawned by the introduction of switches with flexible forwarding capabilities, and routing methods for irregular networks. This paper provides a simple mechanism for cost optimal construction of irregular networks based on three criterions: (1) Distributing the nodes over the switches in such a way that the resulting link utilisation is balanced (when assuming a uniform traffic distribution); (2) using as small number of switches as possible; (3) using the smallest possible switches. This is assuming no parallel links as these will add additional ports to the switch without helping the connectivity. Assuming switch scalability up to a certain limit, our construction gives cost optimal network topologies up to this limit. This provides insight into the influence of the switch size on network cost. Beyond the limitation in switch size we indicate strategies for further scalability that may also be used for the case of having only fixed sized switches.

Computacion De Altas Prestaciones Actas De Las Xv Jornadas De Paralelismo Almeria 15 16 Y 17 De Septiembre De 2004 2004 Isbn 84 8240 714 7 Pags 283 288, 2004

Nowadays, massively parallel computing systems are being built with thousands of nodes. This huge... more Nowadays, massively parallel computing systems are being built with thousands of nodes. This huge number of nodes significantly affects the probability of failure. Thus, it is critical to keep these systems running even in the presence of failures. The interconnection network plays a key role in the performance achieved by these systems, since failures in the interconnection network may isolate a large fraction of the machine containing many healthy nodes. In this paper we present a methodology to design fault-tolerant routing algorithms for regular direct interconnection networks. It supports fully adaptive routing, does not degrade performance in the absence of faults, and supports a reasonably large number of faults without significantly degrading performance. The methodology is mainly based on the selection of an intermediate node (if needed) for each source-destination pair. Packets are adaptively routed to the intermediate node and, at this node, without being ejected, they are adaptively forwarded to their destination. In order to allow deadlock-free minimal adaptive routing, the methodology requires only one additional virtual channel (for a total of three), even for tori. Evaluation results for a 4 × 4 × 4 torus network show that the methodology is 5-fault tolerant. Indeed, for up to 14 link failures, the percentage of fault combinations supported is higher than 91.3%. Additionally, network throughput degrades by less than 10% when injecting three random link faults without disabling any node. In contrast, a mechanism similar to the one proposed in the BlueGene/L, that disables some network planes, would strongly degrade network throughput by 79%.

Scandinavian Conference on Image Analysis, 2001

To effectively utilize the growing number of digital me- dia sources in todays electronic world, ... more To effectively utilize the growing number of digital me- dia sources in todays electronic world, easy to use com- putational flexible content-based access is a necessity. In contrast, content-based access must often be based on in- terpreting results from low-level computationally expensive feature extraction algorithms. We propose a new approach based on dynamic object-oriented Bayesian networks for encapsulating low-level feature

Real-time content-based access to live video data requires content analysis applications that are... more Real-time content-based access to live video data requires content analysis applications that are able to process the video data at least as fast as the video data is made available to the application and with an acceptable error rate. Statements as this express quality of service (QoS) requirements to the application. In order to provide some level of control of the QoS provided, the video content analysis application must be scalable and resource aware so that requirements of timeliness and accuracy can be met by allocating additional processing resources. In this paper we present a general architecture of video content analysis applications including a model for specifying requirements of timeliness and accuracy. The salient features of the architecture include its combination of probabilistic knowledge-based media content analysis with QoS and distributed resource management to handle QoS requirements, and its independent scalability at multiple logical levels of distribution. We also present experimental results with an algorithm for QoS-aware selection of configurations of feature extractor and classification algorithms that can be used to balance requirements of timeliness and accuracy against available processing resources. Experiments with an implementation of a real-time motion vector based object-tracking application, demonstrate the scalability of the architecture.

Proceedings of the First International Workshop on Network Aware Data Management, Nov 14, 2011

ABSTRACT End-point hotspots can cause major slowdowns in interconnection networks due to head-of-... more ABSTRACT End-point hotspots can cause major slowdowns in interconnection networks due to head-of-line blocking and congestion. Therefore, avoiding congestion is important to ensure high performance for the network traffic. It is especially important in situations where permanent congestion, which results in permanent slowdown, can occur. Permanent congestion occurs when traffic has been moved away from a failed link, when multiple jobs run on the same system, and compete for network resources, or when a system is not balanced for the application that runs on it. In this paper we suggest a mechanism for dynamic allocation of virtual lanes and live optimization of the distribution of flows between the allocated virtual lanes. The purpose is to alleviate the negative effect of permanent congestion by separating network flows into slow lane and fast lane traffic. Flows destined for a end-point hot-spot is placed in the slow lane and all other flows are placed in the fast lane. Consequently, the flows in the fast lane are unaffected by the head-of-line blocking created by the hot-spot traffic. We demonstrate the feasibility of this approach using a modified version of OFED and OpenSM with fat-tree routing on a small InfiniBand cluster. Our experiments show an increase in throughput ranging from 150% to 468% compared to the conventional fat-tree algorithm in OFED.

The SCI standard describes the interconnection of computing nodes by a ring structure. Although t... more The SCI standard describes the interconnection of computing nodes by a ring structure. Although the SCIringlet has the bene t of facilitating small con gurations without requiring switches, there are weaknesses with respect to scalability and fault tolerance. We propose to use a w ormhole routed switching fabric, such as the Rcube, as a building block for constructing scalable and fault tolerant switched SCI systems. A thorough discussion is given on the implementation of the SCI to HIC bridges with respect to bu er requirements, deadlock a voidance, scalability and fault tolerance.

Due to the limited processing resources available on a typical host, monolithic multimedia conten... more Due to the limited processing resources available on a typical host, monolithic multimedia content analysis applications are often restricted to simple content analysis tasks, covering a small number of media streams. This limitation on processing resources can often be reduced by parallelizing and distributing an application, utilizing the processing resources on several hosts. However, multimedia content analysis applications consist of multiple logical levels, such as streaming, filtering, feature extraction, and classification. This complexity makes parallelization and distribution a difficult task, as each logical level may require special purpose techniques. In this paper we propose a component-based framework where each logical level can be parallelized and distributed independently. Consequently, the available processing resources can be focused on the processing bottlenecks at hand. An event notification service based interaction mechanism is a key factor for achieving this flexible parallelization and distribution. Experiments demonstrate the scalability of a real-time motion vector based object tracking application implemented in the framework. Authors are listed alphabetically The DMJ project is funded by the Norwegian Research Council under grant no. 126103/431

We study the problem of routing based fault tolerance in SCI-networks. The paperprovides a two st... more We study the problem of routing based fault tolerance in SCI-networks. The paperprovides a two step algorithm to route around faulty regions. The rst phase consists of nding new paths for the packets that should have traversed the faulty region. The second phase, which is the challenging one, consists of modifying the routing strategy, so that the resulting con guration is free from deadlocks. The strategy has been evaluated through simulation of node-faults in a 12 8 torus, using the SCI-model for OPNET.

In this paper we present a parameterized SCI node model implemented in OPNET Modeler. It contains... more In this paper we present a parameterized SCI node model implemented in OPNET Modeler. It contains a full implementation of the go-bit based flow control, as well as the A-B aging scheme of the retry protocol. The simulation environment offers a highly graphical user interface, allowing e.g. the formation of ringlets and larger topologies through drag and deop functionality. A standard distribution of OPNET contains models of most standard communication technologies like Ethernet, FDDI, ATM, etc. This facilitates easy integration of our node with models of other technologies, enabling simulation of heterogeneous systems with very limited additional development.

The ability to tolerate one failing link in communication networks is sufficient for many practic... more The ability to tolerate one failing link in communication networks is sufficient for many practical purposes. One-fault tolerance can also be achieved at much lower cost than methods that can guarantee tolerance of multiple faults. We consider wormhole routed meshes with two different routing algorithms - dimension order and positive first - and for each of these we propose two

Proceedings of the 17th International Symposium on Parallel and Distributed Processing, 2003

In recent years, we have seen an increased interest in irregular network topologies. This has bee... more In recent years, we have seen an increased interest in irregular network topologies. This has been spawned by the introduction of switches with flexible forwarding capabilities, and routing methods for irregular networks. This paper provides a simple mechanism for cost optimal construction of irregular networks based on three criterions: (1) Distributing the nodes over the switches in such a way that the resulting link utilisation is balanced (when assuming a uniform traffic distribution); (2) using as small number of switches as possible; (3) using the smallest possible switches. This is assuming no parallel links as these will add additional ports to the switch without helping the connectivity. Assuming switch scalability up to a certain limit, our construction gives cost optimal network topologies up to this limit. This provides insight into the influence of the switch size on network cost. Beyond the limitation in switch size we indicate strategies for further scalability that may also be used for the case of having only fixed sized switches.

Computacion De Altas Prestaciones Actas De Las Xv Jornadas De Paralelismo Almeria 15 16 Y 17 De Septiembre De 2004 2004 Isbn 84 8240 714 7 Pags 283 288, 2004

Nowadays, massively parallel computing systems are being built with thousands of nodes. This huge... more Nowadays, massively parallel computing systems are being built with thousands of nodes. This huge number of nodes significantly affects the probability of failure. Thus, it is critical to keep these systems running even in the presence of failures. The interconnection network plays a key role in the performance achieved by these systems, since failures in the interconnection network may isolate a large fraction of the machine containing many healthy nodes. In this paper we present a methodology to design fault-tolerant routing algorithms for regular direct interconnection networks. It supports fully adaptive routing, does not degrade performance in the absence of faults, and supports a reasonably large number of faults without significantly degrading performance. The methodology is mainly based on the selection of an intermediate node (if needed) for each source-destination pair. Packets are adaptively routed to the intermediate node and, at this node, without being ejected, they are adaptively forwarded to their destination. In order to allow deadlock-free minimal adaptive routing, the methodology requires only one additional virtual channel (for a total of three), even for tori. Evaluation results for a 4 × 4 × 4 torus network show that the methodology is 5-fault tolerant. Indeed, for up to 14 link failures, the percentage of fault combinations supported is higher than 91.3%. Additionally, network throughput degrades by less than 10% when injecting three random link faults without disabling any node. In contrast, a mechanism similar to the one proposed in the BlueGene/L, that disables some network planes, would strongly degrade network throughput by 79%.

Scandinavian Conference on Image Analysis, 2001

To effectively utilize the growing number of digital me- dia sources in todays electronic world, ... more To effectively utilize the growing number of digital me- dia sources in todays electronic world, easy to use com- putational flexible content-based access is a necessity. In contrast, content-based access must often be based on in- terpreting results from low-level computationally expensive feature extraction algorithms. We propose a new approach based on dynamic object-oriented Bayesian networks for encapsulating low-level feature

Real-time content-based access to live video data requires content analysis applications that are... more Real-time content-based access to live video data requires content analysis applications that are able to process the video data at least as fast as the video data is made available to the application and with an acceptable error rate. Statements as this express quality of service (QoS) requirements to the application. In order to provide some level of control of the QoS provided, the video content analysis application must be scalable and resource aware so that requirements of timeliness and accuracy can be met by allocating additional processing resources. In this paper we present a general architecture of video content analysis applications including a model for specifying requirements of timeliness and accuracy. The salient features of the architecture include its combination of probabilistic knowledge-based media content analysis with QoS and distributed resource management to handle QoS requirements, and its independent scalability at multiple logical levels of distribution. We also present experimental results with an algorithm for QoS-aware selection of configurations of feature extractor and classification algorithms that can be used to balance requirements of timeliness and accuracy against available processing resources. Experiments with an implementation of a real-time motion vector based object-tracking application, demonstrate the scalability of the architecture.

Proceedings of the First International Workshop on Network Aware Data Management, Nov 14, 2011

ABSTRACT End-point hotspots can cause major slowdowns in interconnection networks due to head-of-... more ABSTRACT End-point hotspots can cause major slowdowns in interconnection networks due to head-of-line blocking and congestion. Therefore, avoiding congestion is important to ensure high performance for the network traffic. It is especially important in situations where permanent congestion, which results in permanent slowdown, can occur. Permanent congestion occurs when traffic has been moved away from a failed link, when multiple jobs run on the same system, and compete for network resources, or when a system is not balanced for the application that runs on it. In this paper we suggest a mechanism for dynamic allocation of virtual lanes and live optimization of the distribution of flows between the allocated virtual lanes. The purpose is to alleviate the negative effect of permanent congestion by separating network flows into slow lane and fast lane traffic. Flows destined for a end-point hot-spot is placed in the slow lane and all other flows are placed in the fast lane. Consequently, the flows in the fast lane are unaffected by the head-of-line blocking created by the hot-spot traffic. We demonstrate the feasibility of this approach using a modified version of OFED and OpenSM with fat-tree routing on a small InfiniBand cluster. Our experiments show an increase in throughput ranging from 150% to 468% compared to the conventional fat-tree algorithm in OFED.

The SCI standard describes the interconnection of computing nodes by a ring structure. Although t... more The SCI standard describes the interconnection of computing nodes by a ring structure. Although the SCIringlet has the bene t of facilitating small con gurations without requiring switches, there are weaknesses with respect to scalability and fault tolerance. We propose to use a w ormhole routed switching fabric, such as the Rcube, as a building block for constructing scalable and fault tolerant switched SCI systems. A thorough discussion is given on the implementation of the SCI to HIC bridges with respect to bu er requirements, deadlock a voidance, scalability and fault tolerance.

Due to the limited processing resources available on a typical host, monolithic multimedia conten... more Due to the limited processing resources available on a typical host, monolithic multimedia content analysis applications are often restricted to simple content analysis tasks, covering a small number of media streams. This limitation on processing resources can often be reduced by parallelizing and distributing an application, utilizing the processing resources on several hosts. However, multimedia content analysis applications consist of multiple logical levels, such as streaming, filtering, feature extraction, and classification. This complexity makes parallelization and distribution a difficult task, as each logical level may require special purpose techniques. In this paper we propose a component-based framework where each logical level can be parallelized and distributed independently. Consequently, the available processing resources can be focused on the processing bottlenecks at hand. An event notification service based interaction mechanism is a key factor for achieving this flexible parallelization and distribution. Experiments demonstrate the scalability of a real-time motion vector based object tracking application implemented in the framework. Authors are listed alphabetically The DMJ project is funded by the Norwegian Research Council under grant no. 126103/431

We study the problem of routing based fault tolerance in SCI-networks. The paperprovides a two st... more We study the problem of routing based fault tolerance in SCI-networks. The paperprovides a two step algorithm to route around faulty regions. The rst phase consists of nding new paths for the packets that should have traversed the faulty region. The second phase, which is the challenging one, consists of modifying the routing strategy, so that the resulting con guration is free from deadlocks. The strategy has been evaluated through simulation of node-faults in a 12 8 torus, using the SCI-model for OPNET.