Under the topic of deployment with machine-learning, there are a lot of things to consider and a lot of different options that will provide you with a different result. Firstly, there are a lot of standard VPS and semi-shared hosts that you could go with for deploying your models. These are usually not great options for a beginner but allow for more freedom in what applications are used and how the file-system is arranged. Another option is, of course, a deployment solution like AWS or Google Cloud. While these are both great options, neither of them will give you the price-to-performance ratio and convenience of hosting a server yourself. Although there are of course drawbacks to hosting a server yourself, it is certainly a great option if you happen to have good and reliable internet and an extra computer lying around.
Setup
When you’re running your own server, something very important to consider is what it is actually going to run on. You might not have a full-blown server sitting in your closet like many computing maniacs, but that is fine. For most applications, it’s unlikely that you’re going to have a need for 128GB of memory and a ridiculously high core count processor. For example, if you wish to only deploy a few endpoints, maybe serve some static files, you could go with essentially any computer with more than 1GB of memory.
That being said, another thing you’ll want to consider is heat. How hot does it get where you live, and do you have centralized air-conditioning? It’s important to keep your server components cool, and server processors generate a lot of heat and will require a lot more environmental cooling than your average computer. As a result, you could even have a server but not want to use it because of the heat in your environment.
Another thing that you might want to consider is power consumption. After all, if you’re not using all of your server’s power, but running a high TDP processor, it might not even be cost-effective to host it yourself — you might end up paying more for electricity than you bargained for.
A great suggestion that I have when you don’t need all that much power is to use a laptop. Many people have an old I-7 or I-5 laptop with about 4GB of memory sitting around somewhere, and these are a great option for a simple server that can perform a lot of operations for incredibly cheap. Laptop processors have dramatically lower TDP compared to desktop processors, and as a result, they don’t generate nearly as much heat or use very much power.
#python #programming #servers #linux #deployment
