Given a set of demands between pairs of nodes, we examine the traffic engineering problem of flow... more Given a set of demands between pairs of nodes, we examine the traffic engineering problem of flow routing and fair bandwidth allocation where flows can be split to multiple paths (e.g., MPLS tunnels). This paper presents an algorithm for finding an optimal and global per-commodity max-min fair rate vector in a polynomial number of steps. In addition, we present a fast and novel distributed algorithm where each source router can find the routing and the fair rate allocation for its commodities while keeping the locally optimal max-min fair allocation criteria. The distributed algorithm is a fully polynomial epsilon-approximation (FPTAS) algorithm and is based on a primal-dual alternation technique. We implemented these algorithms to demon-strate its correctness, efficiency, and accuracy. I.
All in-text references underlined in blue are linked to publications on ResearchGate, letting you... more All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.
This volume contains the 17 papers selected out of the contributions submitted to the academic tr... more This volume contains the 17 papers selected out of the contributions submitted to the academic track of SYSTOR 2009 --- the Israeli Experimental Systems Conference, which was held from May 4 to 6, 2009, at IBM Haifa Research Lab, Haifa, Israel. In addition to the papers included in this volume, the conference program also featured keynotes by Professor Marc Snir, UIUC, and Alain Azagury, IBM/XIV Business Executive and member of IBM Academy; an invited talk by Michael Kagan, CTO of Mellanox Technologies, as well as 7 presentations by industry leaders showcasing recent technological advances in the systems and storage area. The student attendees had an opportunity to present their work and interact with senior researchers at the conference's poster session. The conference was closed with a panel "What is systems research about and is it relevant?" moderated by Dror Feitelson and Muli Ben-Yehuda. The goal of SYSTOR is to promote systems research and foster ties between th...
2018 IEEE 11th Conference on Service-Oriented Computing and Applications (SOCA), 2018
The growing smartphone user base has enabled new, paid, mobile crowdsourcing marketplaces, where ... more The growing smartphone user base has enabled new, paid, mobile crowdsourcing marketplaces, where individuals are paid to perform tasks using their mobile phones as they move around in their day-to-day lives. The massive crowdsourcing system serves task requesters who can proactively order data collection tasks from registered workers. The system should allocate incoming tasks to the better workers and still keep the costs of marketplace cloud backend low. We have built an online scoring mechanism that suits large-scale systems where each worker is evaluated continuously according to several parameters and an incoming task is allocated to top-grade workers. The quality of the scoring algorithm is determined by the average time a task remains in the system. The research evaluates the quality of the simple scoring algorithm that considers a worker's queue size at that instant (JSQ) and compares it with a simpler method that considers a worker's average queue length (AQL) with h...
Given a set of demands between pairs of nodes, we examine the traffic engineering problem of flow... more Given a set of demands between pairs of nodes, we examine the traffic engineering problem of flow routing and fair bandwidth allocation where flows can be split to multiple paths (e.g., MPLS tunnels). This paper presents an algorithm for finding an optimal and global per-commodity max-min fair rate vector in a polynomial number of steps. In addition, we present a fast and novel distributed algorithm where each source router can find the routing and the fair rate allocation for its commodities while keeping the locally optimal max-min fair allocation criteria. The distributed algorithm is a fully polynomial epsilon-approximation (FPTAS) algorithm and is based on a primal-dual alternation technique. We implemented these algorithms to demon-strate its correctness, efficiency, and accuracy. I.
All in-text references underlined in blue are linked to publications on ResearchGate, letting you... more All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.
This volume contains the 17 papers selected out of the contributions submitted to the academic tr... more This volume contains the 17 papers selected out of the contributions submitted to the academic track of SYSTOR 2009 --- the Israeli Experimental Systems Conference, which was held from May 4 to 6, 2009, at IBM Haifa Research Lab, Haifa, Israel. In addition to the papers included in this volume, the conference program also featured keynotes by Professor Marc Snir, UIUC, and Alain Azagury, IBM/XIV Business Executive and member of IBM Academy; an invited talk by Michael Kagan, CTO of Mellanox Technologies, as well as 7 presentations by industry leaders showcasing recent technological advances in the systems and storage area. The student attendees had an opportunity to present their work and interact with senior researchers at the conference's poster session. The conference was closed with a panel "What is systems research about and is it relevant?" moderated by Dror Feitelson and Muli Ben-Yehuda. The goal of SYSTOR is to promote systems research and foster ties between th...
2018 IEEE 11th Conference on Service-Oriented Computing and Applications (SOCA), 2018
The growing smartphone user base has enabled new, paid, mobile crowdsourcing marketplaces, where ... more The growing smartphone user base has enabled new, paid, mobile crowdsourcing marketplaces, where individuals are paid to perform tasks using their mobile phones as they move around in their day-to-day lives. The massive crowdsourcing system serves task requesters who can proactively order data collection tasks from registered workers. The system should allocate incoming tasks to the better workers and still keep the costs of marketplace cloud backend low. We have built an online scoring mechanism that suits large-scale systems where each worker is evaluated continuously according to several parameters and an incoming task is allocated to top-grade workers. The quality of the scoring algorithm is determined by the average time a task remains in the system. The research evaluates the quality of the simple scoring algorithm that considers a worker's queue size at that instant (JSQ) and compares it with a simpler method that considers a worker's average queue length (AQL) with h...
Uploads
Papers by Miriam Allalouf