Prediction-based load shedding for burst data streams

Abstract

detail records are generated by telephone exchanges at the end of each user session and describe the session in detail. By analyzing streams of CIMs and CDRs, telecommunication companies can develop procedures to improve the number of calls completed successfully, identify potential revenue leakage, identify the most and the least profitable services (through minutes of usage), find out how existing products are consumed by end users, and discover services to be either expanded or limited. They can also detect logical network problems caused by routing errors,

Monitoring high performance data streams in vertical markets: Theory and applications in public safety and healthcare

“…Finally, we have demonstrated applications of these concepts in the public safety and health-care arenas. We are currently designing an advanced load-shedding mechanism based on prediction methods [17]; however, more extensive study is required to provide a comprehensive analysis.…”

“…However, in the load-shedding scenarios, the query engine shedding mechanism continuously monitored all arriving tuples and simply removed those exceeding processing capabilities [17]. Hence, our similarity-based operator performed considerably better when random drop load shedding was employed, and the average number of recognized faces was higher.…”

“…Hence our approach performs considerably better than one without such a mechanism. An analysis of the load shedding algorithm does not fall within the scope of this paper; however, the efficiency of this approach has been discussed in our previous work [17]. The associated performance improvements have been verified in the lab.…”

“…We are currently working on the implementation of advanced load shedding techniques [17] to reduce the cost of processing incoming data and avoid system overload while minimizing the effect on recognition accuracy. These algorithms will operate by determining which video frames are most important to the integrity of the feed and avoiding the discard of tuples that affect the overall image quality.…”

Similarity based join over audio feeds in a multimedia data stream management system

“…New syntax was implemented to employ automatic human identification using speaker recognition by adding the ASR keyword to the using clause; i.e., select * from stream1, stream2 where attr1 similar attr2 with 0.5 using ASR . Moreover, our solution employs load‐shedding techniques to increase DSMS processing capabilities [15]. Our MMDSMS has been deployed on hosts configured with two 2.5 GHz Intel Core* i5 central processing units (CPUs) and 4.0 GB of random access memory (RAM).…”