It’s not a TV…
"If they want to binge . . . then we should let them binge." – Kevin Spacey
The Netflix model certainly proved one thing that the audience wanted more control. When the company was first launched in 1997 by Reed Hastings, the whole idea sounded insane to most of the people. It started as a disc shipping business and the overall trajectory of the company was hardly predetermined; the naysayers even predicted that it was bound to fail!! The company, however took over the roof-top making this venture a great success and capitalized on it.
By its decade mark and with the advent of internet, streaming video services became much more accessible to a common user. The company evolved into an “on-demand” internet streaming media provider,
…show more content…
Earlier, all the developers were expected to coordinate their efforts in integrating the code and then push it to the production, which led to unavoidable delays and a production nightmares. This transition not only increased the reliability from the sub-system failures, but allowed the system to be scaled up across all the tiers - moving all the way up from the persistence layer to the presentation layer.
Embrace the data…
The first challenged faced by Netflix was around choosing the type of database for this new contemplated architecture.
The older system comprised of Oracle relational database engine which was limited by its capacity to handle the vast data. Moreover, in order to maintain an always-available relational database system, they would have needed to build hot-swaps, hot-spares and a multiple failover systems. This would have not only increased the overall complexity of the system, but would have also made it a prohibitively expensive proposition. However, it had an inherent transactional consistency requirement in every scenario and was a time tested
…show more content…
They eventually learnt that neither was the right answer!! In due course, they switched to a more scalable solution called Cassandra. Apache Cassandra is a distributed storage system for managing very large amounts of structured data. This could be spread out across many commodity servers, and provides highly available service with no single point of failure. The system runs on the network of hundreds of nodes spread across the Amazon cloud infrastructure and can also be configured on cheap commodity hardware. But, as with any software component, the reliability goes down with the scale of the process. Cassandra doesn’t support the conventional relational database model, instead it provides clients with a model that supports dynamic control over the layout and format of data stored.
This proved to be a “Silver Bullet” that Netflix was looking for, it was flexible in terms of scale and allowed multi-region replication. To their advantage, it was supported on the Amazon EC2 and had a high read/write throughput, along with the support for multi-region clusters. It supported Amazon availability zones and provided numerous features like online snapshot, restore/rollback to ensure the data integrity and a fail-over redundancy. It could also be configured under different profiles which could handle heavy read/write cycles, batch writes and rapidly increasing