Where the Cloud Can’t Save You

Generals will divide their forces to ensure that their army survives a battle. The cloud takes you to the next level, you can achieve the same by replicating your application, being in multiple places at once to keep a service window always open for your users.

However, if the split forces still report to a single commander, and he gets taken captive, it doesn't matter how many places you are located, none of them will be able to function. The army as a whole is vulnerable at a single point of failure.

A distributed application can operate in several locations throughout a cloud network, but if every instance of that application is connected to the same database, you have the same problem. Your entire system is vulnerable at the point of its database – even on the cloud.

In order for an army to split up and continue on its mission, local commanders must possess the same information as the general and possess the authority to make commands under his authority.

In order for a database to enable the same agility to the entire application, it must be distributed where several copies of that database exist across multiple nodes, each having the same information and ability to operate independently.

Using Helicopters in a Drone World

Relational databases were first used in the 1960s, long before the cloud, or any type of distributed network. They are designed to work on a server and to provide all your data needs from one location.

Their size and speed enable applications worldwide to serve their users with accuracy and speed.

Their challenge is their perpetual monolithic structure. Due to the complexity of how a relational model puts together data, appearing on multiple nodes while constantly replicating to itself increases that complexity to unsustainable at any cost.

Relational models restrict their applications to a single point of failure. Even at above average availability, cloud platforms do have outages. If such an outage were to hit a relational database, the entire application relying on that database will be disabled until a new server can be found.

A Distributed Solution for a Distributed World

Nonrelational databases were developed alongside the cloud a little over a decade ago. Like cloud platforms, they were developed to be distributed from scratch.

Without the need for multiple tables or even schemas, the most common type of nonrelational database, the document database, is natural for a distributed system where multiple copies of your database can sit at the backend of multiple copies of your application.

The best type of database has a master-master structure where each copy can perform both reads and writes to your data. If you have a database cluster of 3 nodes and one goes down, two databases have full ability and authority to keep working. Even if the majority of your nodes go down, you can still provide service to your users.

Provisioning servers on the cloud closest to where your users are reduces the distance from their device to your application, increasing performance. It also reduces load on any one database at any one time.

It's like being on the longest of three lines in the supermarket when an announcement is made: "Will the people at the end of the longest line please move to the front of the shortest?" How awesome is that.

The Age of Expanding Data

Like our universe, data is in a state of constant expansion. There is always more traffic, more volume, even the amount of information you can store in one unit of data is rising.

To manage that, while keeping performance robust in the face of more information to constantly administer, you need a distributed system. If your application relies on one massive database, the cloud can't save you.

The best way to attack the data of tomorrow is with the finest tools of today.

Author

Oren Eini is the CEO of RavenDB, a NoSQL Distributed Database, and RavenDB Cloud, its Managed Cloud Service (DBaaS). Oren is a Microsoft MVP and a DZone Hall of Famer with over 3.5 million views over ten years writing about NoSQL Database Technology, the .NET Ecosystem, and Software Development. He has been blogging for more than 15 years using his alias Ayende Rahien.