Powering real-time analytics on Xfinity using Kudu
Sridhar Alla and Kiran Muglurmath explain how real-time analytics on Comcast Xfinity set-top boxes (STBs) help drive several customer-facing and internal data-science-oriented applications and how Comcast uses Kudu to fill the gaps in batch and real-time storage and computation needs, allowing Comcast to process the high-speed data without the elaborate solutions needed till now.
Talk Title | Powering real-time analytics on Xfinity using Kudu |
Speakers | |
Conference | Strata + Hadoop World |
Conf Tag | Make Data Work |
Location | New York, New York |
Date | September 27-29, 2016 |
URL | Talk Page |
Slides | Talk Slides |
Video | |
Kudu is redefining the big data ecosystem and opening doors to capabilities not available before. Comcast is moving in the direction of adopting Kudu with Impala and Spark for several projects, including real-time processing of events from Xfinity devices. Sridhar Alla and Kiran Muglurmath explain how real-time analytics on Comcast Xfinity set-top boxes (STBs) help drive several customer-facing and internal data-science-oriented applications and how Comcast uses Kudu to fill the gaps in batch and real-time storage and computation needs, allowing Comcast to process the high-speed data without the elaborate solutions needed till now. Sridhar and Kiran showcase the platform Comcast is testing using Kudu: real-time STB events (tunes) are streamed from Kafka to Spark, which updates Kudu tables with high speed (~5,000 eps) while also sessionizing and maintaining state for tens of millions of devices in Kudu. While the Spark platform updates the transactions in real time directly on HDFS, the middle tier accesses Kudu tables (through Impala) to generate subsecond real-time dashboards while still having the power of Hadoop to deliver batch analytics and integrations with other platforms. This is key to the success of the platform as previously Comcast had to rely on variety of multitiered architectures to both provide fast storage and be able to update just like NoSQL engines—but without the slowness caused by several thousand updates per second. Sridhar and Kiran also explore how Comcast stores half-a-trillion events using Kudu and still gets great performance analyzing the data.