Slides: https://www.slideshare.net/DataStax/tuning-speculative-retries-to-fight-latency-michael-figuiere-minh-do-netflix-cassandra-summit-2016 | The Cassandra architecture shines at ensuring a very high availability of data even while nodes are failing or are overloaded. On the other hand, query latency will often rise during these events, especially on the higher percentiles. Many improvements have been made to reduce this effect over the past years. This talk will focus on one in particular: Speculative Retries. Introduced in Cassandra 2.0 on the server side and in the Java Driver 3.0 on the client side, this strategy remains complex to fully understand and to finely tune. This talk will deep dive into theoretical and practical aspects of Speculative Retries, showing the effect of tuning strategies with ad-hoc benchmarks. About the Speakers Michael Figuiere Cloud Platform Engineer, Netflix Michael is a senior software engineer at Netflix where he works on improving the cloud storage infrastructure. He previously worked at Apple and DataStax where he worked for several years on creating Drivers and Developer Tools for Cassandra. At ease with both enterprise applications and lower level technologies, he specializes in distributed architectures and topics such as databases, search engines, and cloud. Minh Do Senior Distributed Engineer, Netflix Minh Do has been working at Netflix for the last several years to run, patch, and troubleshoot Cassandra on both server and client sides, and is also a co-creator of Dynomite project. Prior to Netflix, at Tango, he spearheaded its Big Data pipeline system from the ground using Spark/Hadoop. Before that, at Qualys, he built a distributed queue system that bridges traffics between all major components. He has passion in distributed system, machine learning/deep learning, and data storages.