BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:THEMIS: Fairness in Federated Stream Processing under Overload - D r Eva Kalyvianaki - The Computer Laboratory\, Cambridge. DTSTART:20180124T161500Z DTEND:20180124T171500Z UID:TALK98437@talks.cam.ac.uk CONTACT:David Greaves DESCRIPTION:Federated stream processing systems\, which utilise nodes from multiple\nindependent domains\, can be found increasingly in multi-provid er cloud\ndeployments\, internet-of-things systems\, collaborative sensing \napplications and large-scale grid systems. To pool resources from severa l\nsites and take advantage of local processing\, submitted queries are sp lit\ninto query fragments\, which are executed collaboratively by differen t\nsites. When supporting many concurrent users\, however\, queries may ex haust\navailable processing resources\, thus requiring constant load shedd ing.\nGiven that individual sites have autonomy over how they allocate que ry\nfragments on their nodes\, it is an open challenge how to ensure globa l\nfairness on processing quality experienced by queries in a federated\ns cenario.\n\nIn this talk I will describe THEMIS\, a federated stream proce ssing system\nfor resource-starved\, multi-site deployments. It executes q ueries in a\nglobally fair fashion and provides users with constant feedba ck on the\nexperienced processing quality for their queries. THEMIS associ ates stream\ndata with its source information content (SIC)\, a metric tha t quantifies\nthe contribution of that data towards the query result\, bas ed on the\namount of source data use to generate it. We provide the THEMIS \ndistributed load shedding algorithm that balances the SIC values of resu lt\ndata. Our evaluation shows that the THEMIS algorithm yields balanced S IC\nvalues across queries\, as measured by Jain's Fairness Index. Our appr oach\nalso incurs a low execution time overhead.\n LOCATION:Lecture Theatre 1\, Computer Laboratory END:VEVENT END:VCALENDAR