· · · Institutional Repository

This thesis presents the design and implementation of a Chip-Multiprocessor (CMP) targeted at streaming applications(e.g. MPEG, MP3). Streaming applications are applications which can be split into several distinct stages working on data elements in a pipelined fashion. We propose a distributed-memory array (MEP- MAS), where the cores communicate via message-passing, optimizing the throughput. Application tasks are dynamically scheduled by a hardware scheduler taking the consumer-producer locality into ac- count, thereby minimizing the communication overhead. The array is evaluated in terms of performance, scalability and predictability as a function of varied input stream sizes, multiple pipelines, number of pipeline stages and traffic volume. The array is configured as a 4 by 5 mesh and has reached speedups as high as 3.6x for a 4-stage pipeline and 13.4x for a 16-stage pipeline. Our experiments have highlighted the need for a balanced workload in order to optimize the performance. Furthermore, it is shown that MEP-MAS is scalable as the speedup and throughput almost linearly increases with the number of added pipelines. The speedup has increased from 3.6x to 13.5x and the throughput from 17k data elements per second to 65k data elements per second. Increasing the traffic volume in the network marginally affects the speedup (-1.9%). Finally, increasing the traffic volume can cause a high deviation in arrival times between two subsequent data blocks in the pipeline of up to 8%.

This thesis presents the design and implementation of a Chip-Multiprocessor (CMP) targeted at streaming applications(e.g. MPEG, MP3). Streaming applications are applications which can be split into several distinct stages working on data elements in a pipelined fashion. We propose a distributed-memory array (MEP-MAS), where the cores communicate via message-passing, optimizing the throughput. Application tasks are dynamically scheduled by a hardware scheduler taking the consumer-producer locality into account, thereby minimizing the communication overhead. The array is evaluated in terms of performance, scalability and predictability as a function of varied input stream sizes, multiple pipelines, number of pipeline stages and traffic volume. The array is configured as a 4 by 5 mesh and has reached speedups as high as 3.6x for a 4-stage pipeline and 13.4x for a 16-stage pipeline. Our experiments have highlighted the need for a balanced workload in order to optimize the performance. Furthermore, it is shown that MEP-MAS is scalable as the speedup and throughput almost linearly increases with the the number of added pipelines. The speedup has increased from 3.6x to 13.5x and the throughput from 17k data elements per second to 65k data elements per second. Increasing the traffic volume in the network marginally affects the speedup (-1.9%). Finally, increasing the traffic volume can cause a high deviation in arrival times between two subsequent data blocks in the pipeline of up to 8%.

Due to the growing popularity of viewing media over the Internet, content servers are suffering from more and more stress every day. This problem is traditionally solved by enhancing the server infrastructure at the content provider, which is effective but also costly. A more cost effective solution would be to use P2P technology to distribute the media stream in real-time. For this purpose, the Chainsaw algorithm has been proposed, which performs very well in simulations. However, Chainsaw has not been implemented in a real video player yet. We have built our own version of Chainsaw called Kettingzaag, and we have added some improvements and features which make it more resillient to errors, such as multiple description coding. Kettingzaag is put to the test in our own video player called Lumberjack, on the DAS-3 supercomputer in Delft. Our experiments show that the Kettingzaag algorithm performs well for network sizes up to a hundred nodes, and is likely to perform just as well for larger network sizes.

Voltage and Frequency Scaling (VFS) has been shown to reduce energy consumption effectively on system level. Most existing work in this field focused on deadline-constrained applications with finite schedule lengths. However, in typical real-time streaming applications, data processing is constantly activated by infinitely long data streams and operations on successive data instances are overlapped to achieve a tight throughput. This necessitates new VFS policies to perform energy efficient processing. In this thesis, we solve throughput-constrained VFS problems for real-time streaming applications with discrete frequency levels on a heterogeneous multi-processor platform. We propose discrete scaling algorithms for a multi-clock domains platform with local voltage switches per processor and for a single-clock domain platform with a global voltage switch for all processors. We prove NP-hardness for the local VFS problem and maximal open for the global VFS problem. A mixed integer linear program (MILP) is formulated for our local voltage scaling algorithm, while for its global counterpart, a three-stage heuristic incorporating MILP is proposed. Furthermore, two extensions of the proposed voltage scaling policies are presented to handle transition overheads and to include application level Time-Division Multiplexing (TDM) schedulers. Experiments show that for our modem application examples, the discrete local VFS algorithm achieves energy savings close to its continuous counterpart, and local voltage switching is much more beneficial in terms of energy saving than global voltage switching. For example, for our Wireless LAN (WLAN) application example, the continuous local VFS algorithm reduces energy by 29.62%, while the discrete local and global VFS algorithms reduce energy by 28.03% and 16.49%, respectively.

The nanoscale represents a fundamentally new regime for lab-on-a-chip type fluidic systems, because it is the typical length scale across which electrostatic forces are mediated. To understand how ionic liquids and charged objects can be manipulated in nanofluidic devices, a fundamental understanding about the structure of the electrostatic double layer and the mechanisms that govern the effective surface charge is needed. These topics were experimentally investigated using the method of streaming currents through individual well-defined nanochannels. The first part of this Ph. D. work contains measurements of the effective surface charge under various ionic conditions including multivalent ions, for which the effective surface charge was found to change sign. The second part discusses the efficiency of electrical power generation by means of streaming currents, both theoretically and experimentally. From these measurements we also obtained quantitative information on the so-called Stern layer conductance. Finally, we studied the effect of confinement on the pressure driven transport of DNA polymers. Our results suggest new possibilities for DNA separation in nanofluidic devices.

The Internet has become in the last years more and more a means of conveyance for multimedia delivering. Many solutions have been proposed to gain high quality of service for video on demand in client-server environments, with adaptive algorithms that adjust the bit-rate of a video stream depending on the clients available bandwidth. Providing video on demand over decentralized peer-to-peer systems is an active research field. The variable bit-rate environment that characterizes peer-to-peer networks causes significant difficulties to ensure quality of service and playback continuity for video on demand applications. This thesis addresses the challenge of serving high quality video on demand by designing and implementing a multi bit-rate video on demand architecture for peer-to-peer networks. We propose a switching scheme, an encoding methodology and a novel algorithm for multiple bit-rate video streaming over peer-to-peer networks. Identical video content is encoded into three different sets of streams with different average bit-rates. The novel multi bit-rate algorithm will switch dynamically between the three sets of streams depending on the available bandwidth. Through a series of experiments we present the effectiveness of this architecture in fluctuating bandwidth scenarios.

In the online travel environment it is physically and economically not possible to retrieve the hotel room rates in real-time for every customer request. To overcome this problem the hotel rates are cached, but due to the fact that the suppliers will not send notifications of price changes it is a challenging task to keep the cache up to date. Analysis of the problem of how to improve the cache’s performance, in terms of accuracy and coverage, led to the conclusion that there is no single risk-free overall solution, but that the cache should be enhanced in a step-wise manner minimizing the (financial) impact of a faulty enhancement. The Enhancement Cycle is defined as a number of stages that each step-wise enhancement will go through. The system provides a solution for two of those stages. The system measures the cache performance in a correct manner and predicts the expiration time for a hotel rate. The system uses Complex Event Processing to detect the patterns needed to do the measurements. The system has been generalized and is extended with a Domain Specific Language for a low-effort application to other problems. The system is running in production and processes millions of messages a day. It aggregates and compresses the measurement data without loosing expressivity and its predictions improved the cache’s accuracy.

This project aim to determine the macro streaming, the heat input and crystal nucleation induced by ultrasonic irradiation of a vessel containing solutions at different power input and irradiation time. First, experimental determination of the mixing induced by the ultrasound irradiation at different power input was studied using a high speed, high resolution camera. The particle image velocimetry (PIV) technique was used for these studies. The experiments showed that when increasing the power of the ultrasound processor, the velocity increased subsequently. Also, when power was increased, vorticity was increased. After, the heat input given by the sonotrode was studied using high speed camera and liquid crystals which change the color with temperature. The particle image thermometry (PIT) method was used to determined 2D temperature profiles. From the 2D temperature measurement 2D supersaturation profiles was determined. Experiments showed that when temperature increases, supersaturation decreases. The distribution of supersaturation and temperature were assessed at the same moment. Lastly crystal nucleation was performed at power inputs of 50% and 75% and different insonation time from 30 to 120 seconds. The solution used in these experiments was lactose. It was found that when power input increases, the amount of lactose crystals formed was also increasing. With higher power input and increased insonation time, the number of the crystals increased as well as the mean diameters. However there were exception. Above a certain level of insonation time the volume based diameter did not correspondently grow.

Multimedia services have been popping up at tremendous speed in recent years. A large number of these multimedia streaming systems are introduced to the consumer market. Internet Service Providers, Telecommunications Operators, Service/Content Providers, and end users are interested in the mechanisms they use, the Quality-of-Experience they provide (e.g. video/audio quality, audio-video synchronization, communication delay, start-up time, etc.), the resources they need, the system stability, and the service availability. The multimedia streaming systems analyzed in this thesis include IP layer multicast TV (IPTV), Peer-to-Peer TV (P2PTV), Content Delivery Networking (CDN), Peer-to-Peer Video-on-Demand (P2PVoD), Server-to-Client Video Conferencing (IPVC) and Peer-to-Peer Video Conferencing (P2PVC). This thesis aims to study various kinds of popular streaming systems, through analytical models, measurement experiments, and simulations, to reveal their characteristics and performance in different aspects. Based on this research, we can not only better understand the behavior and limitations of existing systems and find out the key parameters that affect their performance, but also investigate the potential problems and predict the system performance for future cases. By comparing the two general streaming content delivery methods (Server-Client and Peer-to-Peer), we gain in-depth insights on “which is better” and “what determines better” for different services and in different scenarios.

The thesis addresses the ever-persistent problem of quality guarantees in wireless data networks. The quality guarantees are especially important for real-time streaming applications. While in fixed links most of the quality issues could be, and are still resolved by over-provisioning, in the world of wireless communications things get complicated. The main reason for wireless links being a bottleneck in terms of both achievable throughput and packet loss and delay guarantees, is the resource limitation that will always be there. Our work showed that while there is no magic solution that can resolve all link quality-related problems in wireless networks, a lot could be done to mitigate them. The first, and the most important, is to make the radio as adaptive to the changes of the link conditions as possible. This can be done through our advanced hybrid rate-control algorithm. The algorithm combines a stable throughput-based solution with a rapid link-quality feedback supplement into a novel controller that has both swift response and stable performance. The second solution, which is dependant on the first one, is the propagation of link status-related information from the link layer to the application. The latter can use this feedback to change its data stream properties to match those of the underlying wireless link. This behaviour proves most useful in cases when the radio cannot compensate for a deterioration of the radio connection such as a significant drop of the signal strength due to obstruction, or too many users competing to use the same medium.

Centralised solutions for Video-on-Demand (VoD) services, which stream pre-recorded video content to multiple clients who start watching at the moments of their own choosing, are not scalable because of the high bandwidth requirements of the central video servers. Peer-to-peer (P2P) techniques which let the clients distribute the video content among themselves, can be used to alleviate this problem. However, such techniques may introduce the problem of free-riding, with some peers in the P2P network not forwarding the video content to others if there is no incentive to do so. When the P2P network contains too many free-riders, an increasing number of the well-behaving peers may not achieve high enough download speeds to maintain an acceptable service. In this paper we propose Give-to-Get, a P2P VoD algorithm which discourages free-riding by letting peers favour uploading to other peers who have proven to be good uploaders. As a consequence, free-riders are only tolerated as long as there is spare capacity in the system. Our simulations show that even if 20% of the peers are free-riders, Give-to-Get continues to provide good performance to the well-behaving peers. In particular, they show that Give-to-Get performs very well for short videos, which dominate the current VoD traffic on the Internet.