Running Netflix on Cassandra in the Cloud • Adrian Cockcroft • GOTO 2012

This presentation was recorded at GOTO Aarhus 2012 http://gotocon.com Adrian Cockcroft - Director of Architecture for the Cloud Systems Team at Netflix @adriancockcroft ABSTRACT Netflix used to be a traditional Datacenter based architecture using a few large Oracle database backends. Now it is one of the largest cloud based architectures, with master copies of all data living in Cassandra. This talk will discuss how we made the transition, how we automated and open sourced Cassandra management for tens of clusters and hundreds of nodes using Priam and Astyanax, backups, archiving and performance and scalability benchmarks. TIMECODES 0:00 Introduction 2:13 Cassandra Service Pattern 4:17 Production Deployment 5:12 High Availability 6:32 Triple Replicated Persistence 8:03 Cassandra Instance Architecture 9:37 Priam - Cassandra Automation 13:37 Astyanax Query Example 14:15 "Traditional" Cassandra Write Data Flows 15:39 Astyanax - Cassandra Write Data Flows 16:39 Data Flows for Multi-Region Writes Token Aware, Consistency Level - Local Quorum 19:34 Cassandra Backup 20:48 Cassandra Explorer 22:05 ETL for Cassandra 24:33 Open Source Projects 30:38 Scalability from 48 to 288 nodes on AWS 32:39 Cassandra on AWS 33:36 Cassandra Disk vs. SSD Benchmark 36:12 Live Demo Workload 37:15 Jmeter Setup 40:31 Next Steps 43:58 The Hacktion 45:18 Implementation (plan) https://twitter.com/gotocon https://www.facebook.com/GOTOConference http://gotocon.com