This article gives you a sneak peek into our internal strategic decision-making process and explains why we decided to do things the hard way and create our own infrastructure including building our own datacentres and developing our own hardware.
Looking around, we can see that various foreign companies specialising in providing cloud-based services choose different approaches. Some choose to rent the infrastructure as a service (e.g. Heroku in AWS), others opted for creating their own infrastructure – such as OVH and other companies. Some companies have changed their approach over the years– RackSpace is leaving their own infrastructure and moves to Amazon, Dropbox, on the other hand, left Amazon in favour of creating their own solution. For these companies, the decision is mainly based on the analysis of costs; simply put, some companies are large enough to benefit from developing their own infrastructure solutions. On the other hand, companies can choose to outsource the “commodity part” and focus on the development of their own services with added value, which is a perfectly valid reasoning.
We faced a similar decision and after long deliberation and planning, we came to a conclusion to build our own infrastructure. In many ways, this approach includes more risks and is definitely more demanding in respect of investments and given the current size of the company the operating costs might seem unreasonably high. For the Central European market, we are using the services of VSHosting and, starting next year, Zerops.io with locations based in Czech Republic. We built our first data centre here and are currently planning to build a second datacentre. Besides the data centre, we use our own means to create our own solution of the entire backbone network infrastructure including strategic parts of optical paths.
What brought us to this decision?
A, the risk factor
The first impetus for leaving rented third party infrastructure is based on the risk factor. We take pride in the quality of our services and believe that the highest quality can be achieved only by being able to manage the operation risks, ideally down to the high voltage power supply. From the point of view of the SLA, the concept of provision of cloud-based, colocation and similar services is a mixture of limited guarantees; in reality, nobody actually provides a guarantee for the full scale of problems caused. For example, I recommend you read through the standard SLA of Amazon (https://aws.amazon.com/compute/sla/), particularly the part saying that in the case of outages with the downtime of 7 and more hours per month, you are eligible for a 30% discount. As a client, you never know whether the provider of datacentre or cloud-based services is doing everything to eliminate all possible operation risks. You do not know whether the providers carry out inspections of all equipment of the datacentre as they should and you have no real way to see if they do. When maintaining the infrastructure yourselves, you can afford to invest into an “unnecessary” redundancy of those items that pose increased operational risk and threaten your business. We have invested in the N+2 redundancy of cooling system; although the risk of an outage of two cooling units within several hours is really small, it has happened to us before. Same reasons apply for the operation of our own backbone network; we are still forced to rely on other operators we are connected to on the Internet as that is the fundamental principle of the Internet, but we can limit the risks by connecting using multiple networks within a dynamic routing.
As mentioned above, for a company of our size to build its own infrastructure is very expensive in respect of the investment as well as operating costs. But as the company grows, building our own infrastructure makes more sense, particularly as we are able to plan the infrastructure and look for optimal solutions. That is why we came up with the idea of building the datacentre ServerPark DC2 which will be constructed on the basis of DC power supply, combination of direct and indirect freecooling with adiabatic (cooling by evaporation of water mist on the heat exchanger) and utilising the concept of our own servers, which means that we can increase the capacity of the datacentre (up to tens of thousands of server units) whit significant savings on the operation due to reduced power consumption. In numerical terms, the cost savings can amount to 40-45% in comparison with using third party infrastructure.
Locations outside Central Europe
As we are planning to expand our business to other countries, e.g. to the United Kingdom, we are going to develop solutions for local infrastructure. Our ambitions do not include a construction of a datacentre outside of Czech Republic (yet), therefore, we need to use third party datacentres in the Central European countries; however, we are going to solve the remaining parts of the infrastructure ourselves. The above mentioned disadvantages (inability to control the datacentre operation) apply in this case, however, in the case of Zerops.io-type services, we will compensate for the disadvantages by running operations in several independent locations. Moreover, Zerops.io is technically prepared for operating on completely flat infrastructure on unified hardware, which further facilitates things.
What are the disadvantages of using our own infrastructure?
The operation of our own infrastructure is very demanding in terms of personnel and knowledge requirements. It is very easy to make a mistake and cause some technical issue simply due to insufficient experience, which is something that does not happen with rented infrastructure of an experienced provider. Beside the development of our own services with added value, we are also creating our own solutions of the “lower” infrastructure (datacentre, network) which increases the costs and causes additional stress. But we know why we are doing it. We want to be a premium provider.
Co-Founder & CEO
VSHosting / SnackHost / Zerops.io