A Framework of Software Measurement
A Framework of Software Measurement
M.H. Halstead's Software Science - a critical examination
ICSE '82 Proceedings of the 6th international conference on Software engineering
An analysis of live streaming workloads on the internet
Proceedings of the 4th ACM SIGCOMM conference on Internet measurement
IEEE Transactions on Software Engineering
Software Science Revisited: A Critical Analysis of the Theory and Its Empirical Support
IEEE Transactions on Software Engineering
MapReduce: simplified data processing on large clusters
Communications of the ACM - 50th anniversary issue: 1958 - 2008
Advanced event processing and notifications in service runtime environments
Proceedings of the second international conference on Distributed event-based systems
Watching television over an IP network
Proceedings of the 8th ACM SIGCOMM conference on Internet measurement
Stateful bulk processing for incremental analytics
Proceedings of the 1st ACM symposium on Cloud computing
Proceedings of the 6th ACM International Conference on Distributed Event-Based Systems
| Hi-index | 0.00 |
The current approach used to obtain official television channel statistics is based on polls combined with specialized reporting hardware. These are deployed only on a small scale and batch processed every 24~hours. With the enhanced capabilities of present-day IPTV set-top-boxes, network operators can track channel popularity and usage patterns with a degree of precision and sophistication not possible with existing methods. One such network operator, Altibox, is the largest provider of IPTV in Norway with a deployment of over 320,000 set top-boxes. By tapping into the high-volume stream of channel zap events sent from these set-top boxes, very accurate viewership can be obtained and presented in near real-time. In this paper, we examine two programming paradigms for implementing applications to compute viewership based on channel zap events. One based on a general-purpose programming language (Java) and the other based on a highly specialized event stream processing language (EPL). An important characteristic of this application is stateful event processing. We are interested in exploring the trade-offs between these two implementations, to determine their suitability for such applications. Specifically, we are interested in the performance trade-off and the program complexity of each implementation. Our results show that a pure Java implementation has a significant edge over EPL in terms of performance. Although, our numbers cannot be used to draw a general conclusion, it seems indicative that an event stream processing engine would suffer more than a general-purpose language as query complexity grows. We conjecture that this is because it is easier to construct custom data structures for the specific problem in a general-purpose language like Java. In terms of program complexity, EPL has a slight edge in all metrics, and a significant edge when event streams can be reused to perform more complex processing, indicating that less effort is necessary to extend functionality.