If you do not want to create a statistical service, such as a particular data dashboard, statistical model, data robot, etc., on your own company PC, you can do it on a computer outside the company. Or a fraction of an external virtual computer. Or in other ways. It may save you time but you would lose internal expertise.
I will assume that you know the meaning of memory, CPU, SSD.
For example on Digital Ocean we can very conveniently create virtual fractions of third-party computers through what they call droplets, a word that personally leaves me perplexed:
In the first column we have RAM, also mentioned in another case. As you observe, in the last row we have the highest-performing choice, but if you do the math it costs less to have a business PC with the same features, partly because usually the business PC also has a dedicated graphics card that allows you to have other microservices as well, for example, for natural language processing, going through language models.
What happens if you take the cheapest option? I tried to replicate it on Replit, a remote code workbench, where we have half a virtual CPU (0.5 vCPU):
just install a library of about 10MB (shiny) to completely saturate RAM. The CPU has a peak of 100% for a long time. It took several minutes, probably with 1 vCPU it would have taken half. However, too much. Clearly 1 vCPU, for a shiny dashboard in operation, therefore post installation, is enough if few users access that microservice.
On Hetzner we can have a virtual private server (VPS), a confusing word in my opinion, which in any case indicates a fraction of a third-party virtual computer. When we go to add a server to a project, in “type” we find the following:
We can also buy it at auction, usually you get better deals. Obviously dedicated vCPUs, rather than shared, cost more.
For most micro and small business use cases, I recommend never choosing less than 8GB RAM, 4 vCPUs. Also consider that free automations on GitHub use, not surprisingly, 2 to 4 vCPUs provided by Microsoft.
On Amazon we have EC2 (and Lightsail), a jungle of options that usually serve medium-sized companies and up, so it makes no sense to discuss this option here, also for other reasons.
I remind you, however, that you can create fractions of virtual computers even on some company PCs, or even on the purchases above, obviously appropriately sized.
Once you have the remote infrastructure, the virtual third-party computer, you need the “delegation” part: running the code, microservice you want. And there are various ways: in a previous article I put together all the ingredients to run it on the dedicated PC, but there is also the approach where you package the code, or rather, standardize it via containers. Which is more typically used by computer scientists.
If you need brief assistance on this case study, we can hear from you in a free call of about 30 minutes.