PhD thesis abstracts
March 2010
PhD thesis abstractsAleksandra KovacevicPeer-to-Peer Location-based Search: Engineering a Novel Peer-to-Peer Overlay NetworkPersonalization of Internet services is a significant feature and exploiting the users' location brings the most value to it. Location-based services have a wide application range - from emergency, tracking, and navigation services to informational and entertainment services. In existing centrally managed solutions, the results of location-based search are often incomplete or outdated. Additional information about the searched object (e.g. the menu, facilities, prices) is usually not available, as such a huge amount of data and frequent updates (e.g. the number of free places in restaurant) would overload the server. In a peer-to-peer solution, each peer is responsible for the information about the object it represents, therefore, updating and publishing information is done directly without a single point of failure. It is available to a wide community to join and publish their services, as peer-to-peer systems operate at low costs. The main goal of this thesis is to prove the feasibility of engineering a peer-to-peer solution for fully retrievable location-based search. Following an engineering approach, we first examine the most used and referred overlays of different types (unstructured, structured, and hybrid). Comparative evaluation identifies the influence of their design decisions on quality aspects such as efficiency, scalability, robustness, and stability. The foundation for the design of our solution is based on the findings from this study. The resulting overlay, Globase.KOM is a structured superpeer-based overlay in the form of a tree enhanced with interconnections. Superpeers are chosen from publicly reachable, static peers with more capacity, spare bandwidth, and good network connectivity. The world projection is divided into rectangular zones, which do not overlap. Each zone is assigned to a superpeer, located inside this zone. It is responsible for all peers in the zone. Superpeers form the tree, which is based on the subset-relation of their zones. Further contribution is the clear methodology for evaluating peer-to-peer search overlays, by defining metrics and various workloads that address all crucial quality properties. Additionally, in order to model realistic workloads, we discuss the difference between user behavior in popular file-sharing applications and VoIP applications such as Skype. As an evaluation tool, we select the simulation framework PeerfactSim.KOM and extend it to support various geographical distribution of peers on a world map and a location-aware churn model. The evaluation results prove the efficiency, good load balancing, scalability, robustness, and stability of the system. Query resolution is significantly faster than in related solutions. Additionally, the location-awareness of the overlay results in an efficient mapping of the logical overlay to the physical underlay which reduced total transmission delay and unnecessary underlay traffic. Although uneven load distribution seems to be an issue due to the tree structure of the overlay, we prove very good load balance due to interconnections and a careful zone formation, which together diminish a higher expected overload of superpeers in the higher tree levels. Our solution scales logarithmically with growing network size. It proves to be robust and stable under simultaneous failures of superpeers in higher tree levels, or in the same branch of the tree, and in the case of frequent querying. The biggest influence on the quality of our solution is the choice of identifier space, its management, and interconnections. A peer identifier contains the information responsibility and its location. That allows smart selection of interconnections and efficient greedy routing. Interconnections enable bypassing of the superpeers in higher levels of the tree and therefore allow equal load distribution among the superpeers. A fast recovery time and small performance variation under extreme churn and critical failures is to be credited to the various maintenance strategies used in combination. The simulation results are confirmed by a prototype and its applicability is shown in the examples of distributed multimedia communication and future peer-to-peer based collaborative applications.
Andreas PetlundImproving latency for interactive, thin-stream applications over reliable transportA large number of network services use IP and reliable transport protocols. For applications with constant pressure of data, loss is handled satisfactorily, even if the application is latency sensitive. For applications with data streams consisting of intermittently sent small packets, however, users experience extreme latencies more frequently. Due to the fact that such thin-stream applications are commonly interactive and time-dependent, increased delay may severely reduce the experienced quality of the service. When TCP is used for thin-stream applications, events of highly increased latency are common, caused by the way retransmissions are handled. Other transport protocols that are deployed in the Internet, like SCTP, model their congestion control and reliability on TCP, as do many frameworks that provide reliability on top of unreliable transport. We have tested several application- and transport layer solutions, and based on our findings, we propose sender-side enhancements that reduce the application-layer latency in a manner that is compatible with unmodified receivers running commodity opeating systems like Linux, *BSD, Mac OS X and Windows. We have implemented the mechanisms as modifications to the Linux kernel, both for TCP and SCTP. The mechanisms are dynamically triggered so that they are only active when the kernel identifies the stream as thin. To evaluate the performance of our modifications, we have conducted a wide range of experiments using replayed thin-stream traces captured from real applications as well as artificially generated thin-stream data patterns. From the experiments, effects on latency, redundancy and fairness were evaluated. The analysis of the performed experiments shows great improvements in latency for thin streams when applying the modifications. Surveys where users evaluate their experience of several applications' quality using the modified transport mechanisms confirmed the improvements seen in the statistical analysis. The positive effects of our modifications were shown to be possible without notable effects on fairness for competing streams. We therefore conclude that it is advisable to handle thin streams separately, using our modifications, when transmitting over reliable protocols to reduce retransmission latency.
Cheng-Hsin HsuEfficient Mobile Multimedia StreamingModern mobile devices have evolved into small computers that can render multimedia streaming content anywhere and anytime. These devices can extend the viewing time of users and provide more business opportunities for service providers. Mobile devices, however, make a challenging platform for providing high-quality multimedia services. The goal of this thesis is to identify these challenges from various aspects, and propose efficient and systematic solutions to solve them. In particular, we study mobile video broadcast networks in which a base station concurrently transmits multiple video streams over a shared air medium to many mobile devices. We propose algorithms to optimize various quality-of-service metrics, including streaming quality, bandwidth efficiency, energy saving, and channel switching delay. We analytically analyze the proposed algorithms, and we evaluate them using numerical methods and simulations. In addition, we implement the algorithms in a real testbed to show their practicality and efficiency. Our analytical, simulation, and experimental results indicate that the proposed algorithms can: (i) maximize energy saving of mobile devices, (ii) maximize bandwidth efficiency of the wireless network, (iii) minimize channel switching delays on mobile devices, and (iv) efficiently support heterogeneous mobile devices. Last, we give network operators guidelines on choosing solutions suitable for their mobile broadcast networks, which allow them to provide millions of mobile users much better viewing experiences, attract more subscribers, and thus increase the revenues.
Zeljko VrbaImplementation and performance aspects of Kahn process networks - An investigation of problem modeling, implementation techniques, and scheduling strategiesThe appearance of commodity multi-core processors has spawned a wide interest in parallel programming, which is widely-regarded as more challenging than sequential programming. Existing distributed processing frameworks like MapReduce and Dryad are intentionally meant for large batch workloads and fail to efficiently support cyclic workloads with deadlines. In this respect, a Kahn Process Networks (KPN) is a model of concurrency that relies exclusively on message passing, and that has some advantages over parallel programming tools in wide use today: simplicity, graphical representation, and determinism. In this thesis, we investigate the applicability of KPNs to implementing general-purpose parallel computations for multi-core machines. In particular, we investigate 1) how KPNs can be used for modeling general-purpose problems; 2) how an efficient KPN run-time can be implemented; 3) what KPN scheduling strategies give good run-time performance. For these purposes, we have developed Nornir, an efficient run-time system for executing KPNs. With Nornir, we show that it is possible to develop a high-performance KPN run-time for multi-core machines. We experimentally demonstrate that problems expressed in the Kahn model resemble very much their sequential implementations, yet perform much better than when expressed in the MapReduce model, which has become widely-recognized as a simple parallel programming model. Lastly, we use Nornir to evaluate several load-balancing methods: static assignment, work-stealing, our improvement of work-stealing, and a method based on graph partitioning. The understanding brought by this evaluation is significant not only in the context of the Kahn model, but also in the more general context of load-balancing (potentially distributed) applications written in message-passing style.
|
||||||||
|
||||||||
|
||||||||
|