As today is one of my days off, I figure I might try my hand at writing a little bit of a more broad scope article answering a larger question. What is a datacenter? I’m going to answer this question based on my own experience having run my own hardware in production for 6 years and not refer to outside sources too much. If I am wrong on anything or you have your own opinion, please leave a comment below and let me know.
A datacenter is a set of computers and associated services that allow for the housing and serving of digital information. I do want to tackle the question, “What is Data?”, but that is beyond me at this time as it would involve a lot of information theory that many smarter people than myself have thought about.
These computers, whether it be one or 1,000, need a few components to be useful.
The first thing they need is electricity and power infrastructure. With larger datacenters this is a massively important component for cost savings or not. Where you place the datacenter matters in this regard. You need UPS systems for power outage. You need generators. You need fuel. You need proper amperage and breakers to handle a large amount of electricity in a small space. 3-phase power is good if you plan to scale up to tens of servers. High voltage also helps increase efficiency. 220 vs 110 for example. You can run 220v straight into server power supplies and this helps your electrical dollar go further, because there is less conversion taking place between the outlet and your computers.
The next thing you need is proper internet infrastructure. This means fiber optics. That’s why I have our offices where they are. We have great, and cheap, fiber optic internet connections. Fiber optic has low latency and high bandwidth available. As you get more data you want to add in redundant internet connections with different providers, and with different physical directions on the line. If a line goes down going west, you need internet to stay on and have another backup connection running east.
The cost of internet plays a huge role here. Fiber can cost $200.00 or $3,000.00 per month for the same thing, depending on your location. Usually business class fiber shares similar prices, and it’s expensive. I had a quote from a company in Maine for $600.00/month for cable internet at 100 Mbps speeds. I get 1 Gigabit fiber at my current location for $160.00/month.
My philosophy as we design our servers and infrastructure is to try and create simple solutions. With simple solutions it sometimes is slower or not as efficient. We gloss over the inefficiencies by leveraging more hardware or higher performance parts where needed. I bring this up as an example in relation to Fiber internet. When you choose internet, cable might be the same speed as the fiber option, but cable internet is inherently different. It is older and uses different architecture. Fiber internet is the best option available and you will have less issues in general. It mitigates lots of other potential issues that could crop up by having a less performant solution, even at the same speed. Temporary outages, glitches, and other things you cannot afford to happen while hosting people’s internet services and applications.
You then need your hardware and software stack for your datacenter. This I term as: Compute, Storage, and Networking. You can architect these three elements in many different ways to achieve a datacenter system that works or doesn’t work. We use all three at all layers of our software and hardware stack. We have custom built Ryzen based servers, with custom storage servers using ZFS for data safety, connected with 10 gigabit networking to reduce the chance of a network bottleneck. I’ll go into the details of our hardware and software stack at a later time.
The next layer up is the IP address. This is your address that gets people access into your datacenter to watch those videos, see the websites, or read this very article. You get a static IP address from your local Internet Service Provider. They aren’t that expensive, but the cost can add up if you don’t do things like have a reverse proxy and load balancing inside your datacenter to move traffic around.
As technology advances, I propose that the datacenter can get smaller. That is what we do at Altha Technology. We are making the datacenter smaller, and wanting to teach people how to run their own micro-datacenters around the United States. We believe that geography matters for philosophical reasons and for legal reasons. Having sovereignty and control over your data is prudent practice.
It is a viable practice today. We do it every day in production using open source tools that are free and can be found on the internet, built by regular people. This is the open source internet and it’s awesome.
Hire us to help your datacenter project.
Owner, Altha Technology