Greg Duffy, Dropcam | AWS Re:Invent 2013
summary
Company processes petabytes per month of inbound video—more than YouTube—using EC2, S3 and DynamoDB with a custom cloud video server called Nexus (Scala/JVM) and an Activity Recognition layer that learns event types and sends push alerts via RabbitMQ/SQS. About 40% of cameras use cloud video recording (7 or 30 days stored); analytics run on EMR; caching proves largely useless because most clips are viewed zero or one time. Funding and custom hardware scale the system to billions of events per year.
AI-written summary · about
chapters (10)
- 0:00Introduction to Drop Cam
- 1:12Product and company overview
- 2:13Cloud infrastructure and hosting
- 3:16Scale of video processing
- 3:57Company origin story
- 5:00Hardware development and funding
- 6:20Tech stack deep dive
- 8:27Activity recognition and alerts
- 9:29Public cameras and caching
- 10:12Future plans and growth
