The reason you get asked what your actual use case is, is because the answer to that is required before one can make any kind of sensible suggestions.
(I do suspect mapleLC won't even read this post fully, because it's too long, and doesn't answer their question. I guess this, too, will be classified as "unhelpful bullshit".)
For example, a single-purpose server, for example a database server in a rack, can definitely benefit from good hardware IPMI support, preferably via a dedicated ethernet (RJ45) port.
A server that hosts virtual machines, say one or more Apache/Nginx instances for web stuff, doesn't really need IPMI, as the host OS is basically your management interface. In most cases, you'll run very tightly managed OS, typically Linux, on the host, with focus on security (paranoia is good) and userspace tools basically centered around container management.
A storage server (NAS) is a completely different beast, closer to a database, except that access to the storage server should be physically limited. (Compare to a database, which typically servers requests from other hosts in the same site.)
An edge or frontend server is one that bridges the physically secure site network and external networks, for example in computing clusters. They vary from load balancers (which tend to be more like appliances, closer to say routers than servers, nowadays, although the ones in hardware-secure sites can include TLS encryption and thus reduce the server load significantly using dedicated encryption hardware – the physical network between the balancer and the servers then having plain-text data streams, and thus needs to be physically secured) to openvpn and similar services, which security-wise bridges various domains, and whose correct/secure configuration requires some expertise and experience.
DNS servers don't need much storage, but they need to be fast and secure, because they can be hit quite hard. Denial-of-service attacks and various cache poisoning attacks are common; and if your DNS is taken down, you're hosed. If you only have a few servers, it makes sense to offsource DNS to a company specialized in DNS management, but then (social) hijacking countermeasures must be carefully designed and observed to avoid difficulties. Forgetting to renew a domain is a particularly common failure that can cost a company a lot.
Email servers are an art on their own. For one, you definitely need a fixed IP address, and be prepared to deal with large email service providers like Google and Microsoft. You need both incoming security, as well as outgoing security to avoid getting blacklisted as a spammer. The email service needs to be tightly coupled to proper DNS service, for both reliability and verifiability. Spam filters, DOS and bot attacks, are commonplace, so surprisingly you'll want quite a bit of memory bandwidth and processor power – not the service itself, but for the filtering capabilities.
Computational cluster nodes are a completely separate beast. In many ways they behave exactly like servers – even down to running suitable queue software as the "server" – but their type and resources depends entirely on what one wants to compute with them. Quantum chemistry requires nodes with lots and lots of RAM and GPGPU-type processing power; classical biophysics or materials physics with millions of atoms parallelizes well to separate nodes (machines) with modest memory needs, and just require as much double-precision floating-power arithmetic power as is available.
Then there are IOT servers. You have the ones that handle local area IOT things, and provide typically a web-based interface accessible both locally and usually via internet (secured by usernames and passwords, using http + TLS, aka HTTPS), that don't require much resources at all. A small Linux SBC is usually plenty, although I'd pick one with at least 4 GB of RAM. Plain Debian/Devuan, Apache or Nginx, OpenSSH, fail2ban, and lots of Linux configuration and integration effort will get you there. Then you have the ones that handle remote IOT thingies, and interface to users' phones and so on. These have no local network, and are pure internet ("cloud") beasts. You could consider them like normal web servers, but the fact that the traffic is usually very short streams (as opposed to transferring multi-megabyte files), makes the hardware requirements a bit odd. They're often virtualized with a load balancer in front, and with a separate database server (for "customer" data), so that they can be scaled easily to cater to the needs.
In short, there is no "server". There is a class of machines often called "servers", which spans a wider range of hardware than ordinary desktop computers do.
If you don't know what you want, then go for "dirt cheap", and find out what is available for "dirt cheap". If you have options, then look up their manufacturers sales' pages, to see what they think they are best used for. If you are left with more than one option, list them here and we can tell you more.
Home
Help
Search
Login
Register
