Dwight J. Makaroff

An evaluation of VBR disk admission algorithms for continuous media file servers

In this paper, we address the problem of choosing a disk admission algorithm for continuous media streams where each stream may have a different bit rate, and more importantly, where the bit rate within a single stream may vary considerably. We evaluate several different Variable Bit Rate (VBR) disk admission control algorithms for continuous media. An algorithm which accepts too few streams under-utilizes the server resources, while an algorithm which accepts too many streams over-utilizes the resources resulting in inadequate service (i.e. missing or delayed data) to the clients. The evaluation process is based on a representative set of video streams encoded in MJPEG. We conclude that one particular algorithm, the VBR simulation algorithm, performs the best among realizable algorithms in terms of system utilization and delivery guarantees and performs close to an optimal algorithm,

Characterizing customer groups for an e-commerce website

In conventional commerce, customer groups with similar interests or behaviours can be observed. Similarly, customers in E-commerce naturally form groups. These groups allow the organization to provide quality of service (QoS) and perform capacity planning. From a system point of view, overall server performance can be improved and resources managed considering customer session behaviour.Previous studies have grouped customers using clustering techniques. Different data metrics have been selected as criteria for grouping, in order to analyze different problems. The limitation for these approaches is that problems areanalyzed separately. In order to manage an E-commerce server well, we must analyze many related problems comprehensively rather than separately. For example, we would like to know what is the impact on resource usage when optimizing revenue. Thus, we must understand the differences and similarities between session groups chosen by different metrics.This paper characterizes customer groups for an E-rental business and compares customer groups created according to different criteria including services requested, navigation pattern and resource usage. A significant finding of this study shows that using each of the three criteria independently yields roughly similar results, since customers looking for similar services tend to have similar navigation pattern as well as similar server resource usage. Thus, it issufficient to group customers in only one of these ways. Grouping customers by services requested is suggested since this method yields relatively better results and is simple to implement.

Design of a variable bit rate continuous media file server for an ATM network

This paper describes the design and implementation of a file server for variable bit rate continuous media. Most continuous media file servers have been designed for constant bit rate streams. We address the problem of building a server where each stream may have a different bit rate and, more importantly, where the bit rate within a single stream may vary considerably. Such servers will become increasingly more important because of Variable Bit Rate compression standards such as is possible with MPEG-2.

Teaching requirements engineering to an unsuspecting audience

One of a Software Engineers most important skills is the ability to define the scope of the problem and ascertain the requirements from general and vague specifications. Teaching this skill is known to be difficult and is made more complex because students are conditioned to expect that this portion of programming projects is already complete. This paper reports on experience in teaching a second year computer science class which exposed the need for requirements engineering and gave students an opportunity to engage in the activity. We found that the student response was bimodal, and while some students met the challenge, more felt betrayed by the experience. We conclude that students gained the requisite knowledge using this approach but that a less traumatic approach may produce better results.

Schemes for implementing buffer sharing in continuous-media systems

Buffer management in continuous-media systems is a frequently studied topic. One of the most interesting recent proposals is the idea of buffer sharing for concurrent streams. As analyzed in [6], by taking advantage of the temporal behaviour of concurrent streams, buffer sharing can lead to a 50% savings in total buffer space. In this paper, we study how to actually implement buffer sharing. To this end, we develop the CES Buffer Sharing scheme that is very efficient to implement, and that permits savings asymptotically very close to the ideal savings predicted by the analysis in [6]. We show that the CES scheme can operate effectively under varying degrees of disk utilizations, and during transition periods when the number of concurrent streams changes. We also demonstrate how the scheme can be further improved, particularly for situations when the number of concurrent streams is small. In ongoing work, we will integrate the proposed scheme into a distributed continuous-media file system which is under development at the University of British Columbia.

Performance Evaluation of Yahoo! S4: A First Look

Processing large data sets has been dominated recently by the map/reduce programming model [1], originally proposed by Google and widely adopted through the Apache Hadoop1 implementation. Over the years, developers have identified weaknesses of processing data sets in batches as in MapReduce and have proposed alternatives. One such alternative is continuous processing of data streams. This is particularly suitable for applications in online analytics, monitoring, financial data processing and fraud detection that require timely processing of data, making the delay introduced by batch processing highly undesirable. This processing paradigm has led to the development of systems such as Yahoo! S4 [2] and Twitter Storm.2 Yahoo! S4 is a general-purpose, distributed and scalable platform that allows programmers to easily develop applications for processing continuous unbounded streams of data. As these frameworks are quite young and new, there is a need to understand their performance for real time applications and find out the existing issues in terms of scalability, execution time and fault tolerance. We did an empirical evaluation of one application on Yahoo! S4 and focused on the performance in terms of scalability, lost events and fault tolerance. Findings of our analyses can be helpful towards understanding the challenges in developing stream-based data intensive computing tools and thus providing a guideline for the future development.

Enabling technology for distributed multimedia applications

In September 1993, the Canadian Institute for Telecommunications Research, in collaboration with the IBM Toronto Laboratory Centre for Advanced Studies, initiated a major project on broadband services. The goal of this major project is to provide the software technologies required for the development of distributed multimedia applications. Of particular interest are “presentational“ applications where multimedia documents, stored in database servers, are retrieved by remote users over a broadband network. Emphasis is placed on efficiency and service flexibility. By efficiency, we mean the ability to support many users and many multimedia documents. By service flexibility, we mean that the application is able to support a wide range of quality-of-service requirements from the users, adapt to changing network conditions, and support multiple document types. The research program consists of six constituent projects: multimedia data management, continuous media file server, quality-of-service negotiation and adaptation, scalable video encoding, synchronization of multimedia data, and project integration. These projects are investigated by a multidisciplinary team from eight institutions across Canada. Multimedia news has been selected as a target application for development, and the results from the various projects have been integrated into a multimedia news prototype. In this paper, the system architecture, research results, and the prototyping effort are presented.

Disk cache performance for distributed systems

The influence of client and server cache sizes and the number of clients on caching performance is studied through trace-driven simulation. The results indicate that the locality of reference in disk block reference patterns allows relatively small caches to reduce significantly the number of disk accesses required. File server cache performance is significantly different from client cache performance owing to the capture of disk block references by the client caches. The major factor influencing overall miss ratio statistics (actual disk reference frequencies) is found to be the maximum of the server cache size and the size of client caches.<<ETX>>

Database server workload characterization in an e-commerce environment

In an e-commerce system, the database server performance is crucial. A dynamic cache is often used to reduce the load on the database server, which reduces the need for scalability. A good understanding of the workload characteristics of the database server in an e-commerce environment is important to the design, tuning, and capacity planning of the database server. We characterize the database server workloads in a benchmark e-commerce system. We focus on the response time, CPU utilization, the database page reference characteristics, and disk I/Os of the database server. We find that using a dynamic cache can substantially reduce the CPU utilization but not always the number of disk I/Os of the database server. In most cases, using a dynamic cache reduces the temporal locality in database page references, but to a smaller degree than that reported in file servers and Web proxies. Interestingly, in certain e-commerce workloads, using a dynamic cache results in better temporal locality.

Intrusion detection systems for community wireless mesh networks

Wireless mesh networks are being increasingly used to provide affordable network connectivity to communities where wired deployment strategies are either not possible or are prohibitively expensive. Unfortunately, computer networks (including mesh networks) are frequently being exploited by increasingly profit-driven and insidious attackers, which can affect their utility for legitimate use. In response to this, a number of countermeasures have been developed, including intrusion detection systems that aim to detect anomalous behaviour caused by attacks. We present a set of socio-technical challenges associated with developing an intrusion detection system for a community wireless mesh network. The attack space on a mesh network is particularly large; we motivate the need for and describe the challenges of adopting an asset-driven approach to managing this space. Finally, we present an initial design of a modular architecture for intrusion detection, highlighting how it addresses the identified challenges.