Björn Skubic
Research Subjects
Björn Skubic has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Lena Wosinska
Department of Electrical Engineering, Chalmers University of Technology, Gothenburg, Sweden
4 shared publications - Jawwad Ahmed
Royal Institute of Technology KTH
2 shared publications - Jiajia Chen
Chalmers University of Technology
2 shared publications - Paolo Monti
Department of Electrical Engineering, Chalmers University of Technology, Gothenburg, Sweden
2 shared publications Ahmad Rostami
Ericsson AB
1 shared publication- Ahmed Ali-Eldin
Chalmers University of Technology and University of Gothenburg, Gothenburg, Sweden
1 shared publication
Latest Publications
- Open Access
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 27-39
RAVAS: Interference-Aware Model Selection and Resource Allocation for Live Edge Video Analytics
Numerous edge applications that rely on video analytics demand precise, low-latency processing of multiple video streams from cameras. When these cameras are mobile, such as when mounted on a car or a robot, the processing load on the shared edge GPU can vary considerably. Provisioning the edge with GPUs for the worst-case load can be expensive and, for many applications, not feasible. In this paper, we introduce RAVAS, a Real-time Adaptive stream Video Analytics System that enables efficient edge GPU sharing for processing streams from various mobile cameras. RAVAS uses Q-Learning to choose between a set of Deep Neural Network (DNN) models with varying accuracy and processing requirements based on the current GPU utilization and workload. RAVAS employs an innovative resource allocation strategy to mitigate interference during concurrent GPU execution. Compared to state-of-the-art approaches, our results show that RAVAS incurs 57% less compute overhead, achieves 41% improvement in latency, and 43% savings in total GPU usage for a single video stream. Processing multiple concurrent video streams results in up to 99% and 40% reductions in latency and overall GPU usage, respectively, while meeting the accuracy constraints.
- 6 December 2023
- Published by Association for Computing Machinery (ACM)
- p. 27-39
Open AccessRAVAS: Interference-Aware Model Selection and Resource Allocation for Live Edge Video Analytics
Numerous edge applications that rely on video analytics demand precise, low-latency processing of multiple video streams from cameras. When these cameras are mobile, such as when mounted on a car or a robot, the processing load on the shared edge GPU can vary considerably. Provisioning the edge with GPUs for the worst-case load can be expensive and, for many applications, not feasible. In this paper, we introduce RAVAS, a Real-time Adaptive stream Video Analytics System that enables efficient edge GPU sharing for processing streams from various mobile cameras. RAVAS uses Q-Learning to choose between a set of Deep Neural Network (DNN) models with varying accuracy and processing requirements based on the current GPU utilization and workload. RAVAS employs an innovative resource allocation strategy to mitigate interference during concurrent GPU execution. Compared to state-of-the-art approaches, our results show that RAVAS incurs 57% less compute overhead, achieves 41% improvement in latency, and 43% savings in total GPU usage for a single video stream. Processing multiple concurrent video streams results in up to 99% and 40% reductions in latency and overall GPU usage, respectively, while meeting the accuracy constraints.
- 8 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 34 (4) , 180-187
Flexible RAN: Combining Dynamic Baseband Split Selection and Reconfigurable Optical Transport to Optimize RAN Performance
The introduction of functional splits in C-RANs brings a tradeoff between radio performance and transport capacity. Higher-layer splits relax transport capacity requirements, whereas radio performance is not guaranteed. Lower-layer splits are beneficial for the radio performance, but they may require a more expensive and high capacity transport network. Facing the challenge of how to deploy 5G RANs in the short-term future, network operators need to find the best functional split options able to accommodate radio performance requirements without incurring excessive transport network costs. This article presents an architecture referred to as F-RAN able to choose the most appropriate split option while considering time-varying radio performance and the availability of transport resources. F-RAN can accommodate these needs by means of an SDN-based orchestration layer and a programmable optical transport network. The performance of F-RAN is benchmarked against a conventional C-RAN architecture in terms of the number of wavelengths and transponders to be deployed. Simulation results confirm the overall benefits of F-RAN in terms of better utilization of transport resources.
- 8 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Network
- Vol. 34 (4) , 180-187
Flexible RAN: Combining Dynamic Baseband Split Selection and Reconfigurable Optical Transport to Optimize RAN Performance
The introduction of functional splits in C-RANs brings a tradeoff between radio performance and transport capacity. Higher-layer splits relax transport capacity requirements, whereas radio performance is not guaranteed. Lower-layer splits are beneficial for the radio performance, but they may require a more expensive and high capacity transport network. Facing the challenge of how to deploy 5G RANs in the short-term future, network operators need to find the best functional split options able to accommodate radio performance requirements without incurring excessive transport network costs. This article presents an architecture referred to as F-RAN able to choose the most appropriate split option while considering time-varying radio performance and the availability of transport resources. F-RAN can accommodate these needs by means of an SDN-based orchestration layer and a programmable optical transport network. The performance of F-RAN is benchmarked against a conventional C-RAN architecture in terms of the number of wavelengths and transponders to be deployed. Simulation results confirm the overall benefits of F-RAN in terms of better utilization of transport resources.
- 1 January 2020
- Published by Springer Nature
- p. 831-848
Introduction to Optical Access Networks
- 29 August 2015
- Published by Springer Nature in Photonic Network Communications
- Vol. 30 (3) , 403-415
On the design of 5G transport networks
- 29 August 2015
- Published by Springer Nature in Photonic Network Communications
- Vol. 30 (3) , 403-415
On the design of 5G transport networks
- 21 April 2015
- Published by Springer Nature in Photonic Network Communications
- Vol. 29 (3) , 322-329
Optical transport and challenges in the Networked Society
- 21 April 2015
- Published by Springer Nature in Photonic Network Communications
- Vol. 29 (3) , 322-329
Optical transport and challenges in the Networked Society
- 9 July 2013
- Published by Springer Nature in Telecommunication Systems
- Vol. 54 (2) , 137-146
Enhanced IPACT: solving the over-granting problem in long-reach EPON
- 9 July 2013
- Published by Springer Nature in Telecommunication Systems
- Vol. 54 (2) , 137-146
Enhanced IPACT: solving the over-granting problem in long-reach EPON
- 4 November 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Magazine
- Vol. 48 (11) , 100-108
Dynamic bandwidth allocation for long-reach PON: overcoming performance degradation
A passive optical network, with its inherent point to multi-point structure, allows for centralized placement of active equipment and possible extension of its boundary towards core networks. This property of the PON can be exploited for node consolidation where multiple central offices are replaced by a single one covering a larger service area. Such node consolidation is being particularly driven by the need for network operational cost saving, and is offering significant challenges to PONs. The degree of node consolidation that can be achieved is limited by the reach of conventional PON systems. In order to achieve a larger degree of node consolidation, an extension of the PON reach, beyond the conventional 20 km, is required. This article addresses the challenges of the dynamic bandwidth allocation, where increased reach results in a degradation of DBA performance and quality of service support. This degradation is a consequence of the increased propagation delay of the DBA messages exchanged between different PON elements. A potential solution to the performance degradation is the introduction of a multi-threaded DBA. In this article, we examine for both Gigabit PON and Ethernet PON, the extent to which DBA performance degradation can be reduced by exploiting multi-threading. It is found that for both standards, multi-threading, if done properly, can be used to mitigate the performance degradation due to the increased reach. To make bandwidth allocation efficient, new schemes for coordinating the multiple threads are required in long reach PON.
- 4 November 2010
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Magazine
- Vol. 48 (11) , 100-108
Dynamic bandwidth allocation for long-reach PON: overcoming performance degradation
A passive optical network, with its inherent point to multi-point structure, allows for centralized placement of active equipment and possible extension of its boundary towards core networks. This property of the PON can be exploited for node consolidation where multiple central offices are replaced by a single one covering a larger service area. Such node consolidation is being particularly driven by the need for network operational cost saving, and is offering significant challenges to PONs. The degree of node consolidation that can be achieved is limited by the reach of conventional PON systems. In order to achieve a larger degree of node consolidation, an extension of the PON reach, beyond the conventional 20 km, is required. This article addresses the challenges of the dynamic bandwidth allocation, where increased reach results in a degradation of DBA performance and quality of service support. This degradation is a consequence of the increased propagation delay of the DBA messages exchanged between different PON elements. A potential solution to the performance degradation is the introduction of a multi-threaded DBA. In this article, we examine for both Gigabit PON and Ethernet PON, the extent to which DBA performance degradation can be reduced by exploiting multi-threading. It is found that for both standards, multi-threading, if done properly, can be used to mitigate the performance degradation due to the increased reach. To make bandwidth allocation efficient, new schemes for coordinating the multiple threads are required in long reach PON.
- 1 January 2010
- Published by Optica Publishing Group
- p. OTuO5
Power Efficiency of Next-Generation Optical Access Architectures
A theoretical model for the power consumption of access networks is presented. The analysis based on this model provides insights into optimal number of lines per equipment and the use of active vs. passive equipment.
- 1 January 2010
- Published by Optica Publishing Group
- p. OTuO5
Power Efficiency of Next-Generation Optical Access Architectures
A theoretical model for the power consumption of access networks is presented. The analysis based on this model provides insights into optimal number of lines per equipment and the use of active vs. passive equipment.
- 1 January 2009
- Published by Optica Publishing Group
- p. TuT1
Challenges and Opportunities for Migration towards 10GPON
Different technical challenges for migration from GPON to 10GPON are discussed and also put into context of standardization. The challenges include how to handle an increased uplink capacity through use of burst-mode receivers and DBA algorithms, extension of reach and wavelength planning for future proofing the fiber infrastructure.
- 1 January 2009
- Published by Optica Publishing Group
- p. TuT1
Challenges and Opportunities for Migration towards 10GPON
Different technical challenges for migration from GPON to 10GPON are discussed and also put into context of standardization. The challenges include how to handle an increased uplink capacity through use of burst-mode receivers and DBA algorithms, extension of reach and wavelength planning for future proofing the fiber infrastructure.