[2/2] About Netflix - Infrastructure, Architecture and AI (Category Engineering)
Continue. story Talk about Netflix and talk about the company’s technical foundations.
Netflix runs entirely in the cloud and uses an advanced distributed architecture. All Netflix server systems are deployed to AWS The company has completed a full transition to the Amazon cloud. 2016 Since then, he has not owned his own data center. Today, Netflix has hundreds and thousands of microservices operating in multiple AWS regions. 230+ million accounts worldwide. Applications are divided by functional services (From user profile and recommendation system to payment system or video directory)They communicate with each other through APIs. The transition to microservices architecture has been underway. 2009–2012 The years when Netflix cut its monolithic system into separate services to increase scalability
Netflix has built a whole set of internal Platform solutions on top of AWSIt makes life easier for its developers. Among them are
- Own orchestration system Titus (internal project, then open source). Titus integrated with AWS (manages EC2 instances)It allows engineers to deploy their services in containers without having to manually operate virtual machines.
- Own CD system Spinnaker (internal project, then open source). It runs on top of Titus and automates the deployment of microservices across all regions. (multi-regional deployments)
- Own Gateway API Zuul (internal project, then open source). In an updated version of Zuul 2 It is built on non-blocking Netty and can withstand huge volumes of connections, acting as the first frontier of scaling.
- Own circuit breaker. Hystrix (internal project, then open source). It is interesting that inside Netflix it was abandoned.
- Chaous engineering practices and tools Chaos Monkey, Simian Army: Latency Monkey, Chaos Kong et al.
Architecture Netflix data distributed and scalable
- Apache Cassandra It is used for high-load online storage, where huge volumes and horizontal scaling are required. On top of various storages, Netflix implemented an abstract layer of KV Storage, unifying access for developers and encapsulating details of data replication To caching frequently requested data (Profiles, lists of recommendations, etc.) Several data centers use their own EVCache service. (overhead) Streaming data is processed through a streaming platform Mantis (internal project, then open source). The platform can filter and aggregate millions of events in real time. For big data batch problems, a combination of solutions is used: Keystone - streaming over Kafka, and Genie Orchestra of tasks in Hadoop/Spark clusters (The latest release in open source was 3 year)
- In 2020Netflix opens source Conductor - his orchestrator of business workflow for microservices
- In 2024 A new version of the orchestrator of data / ML-pipelines Maestro.
While Netflix’s computing services run on AWS, the company built its own to deliver video traffic. CDN Netflix Open Connect. This infrastructure solves the problem of streaming giant amounts of data (Tbps) with optimal quality. Open Connect is a fleet of physical caching applications that Netflix installs in ISP nodes around the world.
When it comes to AI/ML,
- User experience personalized, recommendations, film covers AI/ML optimizes service infrastructure to lower bones AI is used to produce content, as well as, for example, to create trailers
This is why Netflix is considered a technology company.
#Culture #Management #Leadership #Processes #Engineering #Software