High availability in the public cloud
20. April 2020
For many companies, the use of the public cloud means that the direct visibility of data centres, servers and the entire infrastructure in their own company is disappearing. When conveniently clicking in web consoles, it is easy to forget that the public cloud is also based on “completely normal” systems and architectures.
1 server is 1 server
Whether physical or virtual – a system remains a system.
Virtualisation makes it possible to flexibly move an instance from one piece of hardware to another and to separate computing power from storage, which makes it easier to rectify failures. Nevertheless, if one or more underlying components fail, the system is not available.
This basic rule applies not only to virtual machines (VMs), but also to all other cloud-based services such as database services (Database as a Service, DBaaS).
Ensuring high availability in the public cloud
Even in the public cloud, systems and applications should be designed for high availability from the outset. To achieve this, public cloud providers rely on availability zones (AZs) and regions. A region usually comprises several isolated availability zones – comparable to two or more independent data centres at the same location.
Take Azure, for example:
In Microsoft Azure, high availability is achieved by physically distributing virtual machines across different zones. A virtual network is set up at each location with the help of load balancers.
The essential steps:
- Use of a zone-redundant load balancer
- Setting up a front-end subnet
- Setting up a database subnet
- Placing the VMs in three different availability zones
- Configuring zone-redundant SQL DB
- Adding the VMs to the back-end pool of the load balancer
- Deploying the application on VMs for redundancy and high availability
SLA: 99% per month, one Typo?
What is more than recommendable from a technical point of view can also be deduced from the SLA of the public cloud providers. For the operation of a VM in an availability zone, for example, Azure has no availability at all. Only those who distribute their VMs in at least 2 availability zones benefit from an SLA. This is stated as 99.99% per month. However, if you read more closely, you will realise that the consequences of non-fulfilment only really come into play if the availability is much lower. And even then, these are only incurred in the form of “service credits”.
More security through well thought-out architecture
Anyone operating systems in the public cloud should always keep an eye on the physical and logical separation of components. This is the only way to ensure that failures of individual hardware or infrastructure elements do not affect the entire operation.
Stability starts with the design
High availability in the public cloud is not a sure-fire success. It requires careful architectural planning, utilisation of the redundancy mechanisms on offer and regular testing. Those who follow these principles will benefit from a flexible, scalable and robust infrastructure.
