Our Cloud Concept
Our concept is built on three major pillars:
- owning the complete stack
- fault tolerance built into the complete stack, including software
- using open source solutions
Each principle is explained below.
Owning the whole stack
We optimize every part of the stack for our cloud services and avoid unnecessary costs. For example, we take advantage of the cold climate in Mo i Rana to implement an optimized cooling solution. We adhere to the temperature specs of the hardware we use - not what is common practice for colocation data centers.
In the same spirit, the Open Compute Project is about simplifying the data center and removing inefficient designs, for example using rack-local batteries instead of a centralized UPS and avoiding multiple conversions between AC and DC.
We are fans of Open Compute and use OCP hardware to provide our services. This allows us to avoid vendor lock-in while utilizing innovation from many of the largest tech companies in the world.
Data centers are traditionally designed with fault tolerant infrastructure (using a 2N or even 2(N+1) architecture for cooling and power, etc.), so the IT layer above it “doesn’t have to worry about” uptime. However, it is a fact that all hardware may fail -- and statistically will fail at some point. No single server or hard drive should be critical by itself. This means that it is insufficient to deal with fault tolerance solely in the infrastructure layers -- one has to involve the IT layer as well.
That is why, instead of having a single data center with highly redundant infrastructure, our cloud is designed with multiple sites (also known as availability zones) that have independent failure domains. We use software to replicate data and relocate load as needed. This allows the service to stay up even if there are issues with an individual site. The sites are generally designed with N+1 redundancy, making single-site downtime a rare, but not impossible event.
We start out with three sites initially, but have a goal of expanding rapidly. Compared to the traditional architecture, this design enables higher uptime at lower cost.
We are an open source-friendly company. We use a lot of open source software in our platform - notably OpenStack, but also other platforms and libraries. As previously mentioned, we use OCP servers, which is open-sourced hardware designs. We believe that open source in the long run outperforms proprietary systems, both technically and from a cost perspective. Besides, it gives us freedom to modify our technology stack in the future.
More info can be found on this page.