Teezily - one of our customers - presented at the OpenIO Summit '17 a very nice story explaining the reasons for their switch from public cloud to hosted bare metal. This story isn’t unique: Dropbox made a similar choice in the recent past. But it’s great to hear that Teezily saved 400K €/Year on storage infrastructure thanks to OpenIO SDS.
The cloud is flexible (and expensive)
This isn’t news. The pay-as-you-go model is fantastic, especially at the beginning of new projects and for start-ups. It allows you to pay for what you actually use, grow or shrink when needed, and avoid capital expenses.
But this flexibility comes at a cost. At the beginning, the cost is irrelevant: you pay for CPU, RAM, storage, and all other resources by the hour or even by the second. But when you start adding up all these resources and data, and use them more heavily, things change quickly. The bill grows and grows, and you start wondering where you can cut costs and improve efficiency.
Growing bills are an issue especially for frontend applications, those that are open to the internet. Most cloud providers charge relatively small amounts to save data in their premises, or for network traffic between peer servers and storage. But things change dramatically as soon as you start serving data and web traffic to the outside world. You get hammered on transactions, traffic peaks, total bandwidth; and if you reach the limit of your contract or tier, then it gets truly painful.
On-premises is predictable (but not flexible)
It’s not easy to find the ideal solution. The cost of a traditional on-premises infrastructure is more predictable, but you pay all this predictability up front before even starting. And it’s not that flexible; you need to be able to adapt your infrastructure to changing business needs quickly.
If your business is web-based and you need some infrastructure services that require bandwidth, reliability, and speed of access from all around the world, building your own infrastructure is too expensive and is hard to manage in the long term.
Fortunately, there is a third option. It sits in the middle between pure IaaS and private infrastructure, and end users often don’t consider it: dedicated servers.
Several service providers have very powerful infrastructures well connected to the internet, and they can provide an inexpensive, pre-configured Linux server. These usually cost a few dollars a month, but prices are fair, and you still have the flexibility of getting a new server up and running in minutes when needed.
The local connection between servers is usually very good (and free), while the cost of raw bandwidth to the rest of the internet is not very pricey. Some of these providers also have firewall and VPN services that can be easily configured. And all administration can be done via UI, CLI, or API.
With this solution, you can build your own private cloud built on top of a public infrastructure, and pay for it monthly. It's not as flexible as IaaS, but it is much less expensive.
How to save €400K (and more)
The hardware (and the OS) is just the first step. Our customer built his own S3-compatible object store on top of these servers, using OpenIO SDS. By doing so, and also by replacing AWS CloudFront with alternative solutions built with a similar approach, that saved €400K in AWS bills. And this was just in the first year.
The application’s performance has been improved by 10x, depending on the type of processes, increasing satisfaction for both users and developers.
And as they look at their business trends, it’s clear that the gap between AWS’s cost in the medium/long term and this new solution will be much larger going forward.
Yes, $400K doesn't include the cost of redesigning the infrastructure (not the application, because it uses the same protocols), nor the additional management effort required by this infrastructure. But how much does it cost to hire a skilled engineer capable of doing this job?
This company is managing 350TB of capacity and 10 billion objects, served around the world through their web application, with just one administrator.
Do your math and see how much you could save.
Industrializing the good stuff
Do you like this idea? Well, we did this already with OVH and their SoYouStart service in the most granular way, using dedicated ARM servers starting with three nodes, and have already tested up to 48 nodes. If you want to know more about this download the whitepaper with all the benchmarks.
Is on-premises object storage too expensive? No problem, OpenIO SDS is so flexible and easy to use that it can be installed everywhere. It can even be installed on a dedicated server on the internet (either X86 or ARM).
The cost of such an infrastructure is lower than full IaaS or Staas (Storage as a Service), and performance may be higher. Everything is tailored for you, as fast or cheap as you need.
Even if this type of infrastructure is not billed by the minute or the hour, but on a monthly basis, it still maintains most of the advantages, while providing more flexibility at a lower cost.
For us, at OpenIO, this is just another option we offer to our customers. Our software solutions are particularly lightweight and can easily adapt to available resources, providing customers with the best freedom of choice.