Dirt cheap decomissioned servers?

[Nominal Animal]

One of my own firewall projects uses a MikroTik RouterBoard M33G, a dual-core MIPS MT7621A running at 880 MHz with 256 MiB of RAM, using OpenWRT (derivative).  It does also run some custom IOT and network connectivity monitoring stuff, plus Dynamic DNS and OpenVPN.  Cost is about 60€ new, plus an LTE modem unless you have wired Internet connection.

These things are the opposite of typical "servers" –– I like to use the term "appliance" –– exactly because they're so memory-constrained, even though they do do server stuff, and run a server-oriented Linux distribution.  For "reasons", mine does actually have a PCIe SSD (M33G supports up to 2242 size-wise), but typically there are also heavy limitations on storage use, affecting things like semi-permanent logging.

(I do also have more typical routers running OpenWRT, mostly acting as a bridge between 2.4 GHz and 5 GHz WiFi, and my wired untrusted LAN sections.)

Currently under development is a funky little service running on OpenWRT that consumes minimal RAM, and reacts to connection attempts by banning the entire IP address for a specified duration.  That is, you can still poke at the firewall ports, but any packets from that IP address on the internet will not be forwarded to the LAN by the firewall at all.  Access attempt pattern (ports and frequency) will determine the duration.  It's an idea I developed based on years of fail2ban use on non-memory-constrained machines.  I'm looking at wiring kernel netfilter packet logs also, so that direct (non-OpenVPN) connection attempts to specific ports on any LAN address will trigger that also, without having to be promiscuous.

[MK14]

These things are the opposite of typical "servers" –– I like to use the term "appliance" –– exactly because they're so memory-constrained, even though they do do server stuff, and run a server-oriented Linux distribution.

Things are changing.  I think such devices, have become very popular in Japan.  Where they wanted to add (potentially) lots of virtualization (lots of VMs, likes lots of ram to be available, usually).  So, memory capacities (specific device dependent), of 32 GB or even 64 GB, is perfectly doable now.  If I remember, correctly.
Even ones with quite a few cores (at least 8, maybe more), not too far from desk top speeds, despite relatively low energy consumption.
Because they essentially typically use the mobile rather than the desktop, versions of the CPUs.

They can be quite reasonably priced, as well.

The three things they don't do well, are significant expandability, graphics cards or long term, high CPU loads, as their cooling solutions are often not suited for such extended, high load duty-cycles.

But many server uses, are fine with such specifications.

[james_s]

When I last worked with this stuff we were starting to see more and more blade servers, an entire computer on a cartridge that plugged into a rack mounted cage. It was all about higher density, packing more and more physical servers into a rack. They had little storage though, typically just a pair of 2.5" HDDs as I recall. These days everything I deal with is VMs, I don't even know what the actual hardware it runs on is, my current employer relies heavily on AWS.

[mapleLC]

The potential advantages of going for a main-stream, popular/common server type/make, such as Dell or HP.  Is you are more likely to find software/drivers, replacement/upgrade parts (some of which is custom and can't come from other computers), google-able/youtube help, and more likely to find help when/if asking on the appropriate forums, if necessary.

Also bios upgrades (if needed), are more likely to be available, although I think HP get funny and may insist on a service subscription or some nonsense like that.  Which may make you avoid them.

There can be, really cheap (apparently identically spec'ed), less common server makes.  But they can end up biting you on the nose, in various ways.  Such as that their spare parts are harder to get, especially quickly, information can be harder to come by and drivers (software) support, can be much more sketchy.

Servers can be much, much more fussy and fiddly than normal PCs.  Things which would have readily worked on a PC, may not be supported on some/all servers.  They are potentially much less plug and play, than PCs.  There are many gotchas, for the unwary.


Ideally, get just about everything you need, in the original server purchase.  Which should solve compatibility issues.

They are more tricky to upgrade (unless you have lots of experience with them), there can be little physical room to expand them (rack ones), and may need special PCI/cards, to fit in a rack case.

You ideally need to decide if you need/want, certain server features.  Such as remote (network) management ports (BMI/IPMI/etc), SATA and/or SAS disks, hardware raid (differing types), how much ECC ram (it often seems to be cheaper to get the right amount of ram, from the beginning), and other stuff, such as graphics cards (and/or professional computation/AI ones), if required.

They often DON'T include the hard disk caddies, which can be an annoying extra cost.

Dual power-supplies, was for the original high availability use of the servers.  For light home casual use, a single power supply may be fine.

1U (small height) servers, tend to be extremely noisy.  Even if in a room/place, well away from people, just working near it for brief periods, can still be very annoying/distracting.  It literally can sound like a jet-aircraft is taking off in the room (REALLY!).

2U (partly taller) servers, are usually a lot quieter, as they have bigger fans, that don't need to spin up to such high RPMs.  But they can still be quite noisy.

You can now get rather modern tiny/small, powerful, lots of Ethernet connections, servers.  With super-fast M.2 SSDs.

You can get numerous specifications, prices etc, so don't panic.

Here is an example:

I will watch the videos, thank you.

One of the reasons I put in a budget is giving myself room for screw ups and buying peripherals I didn't expect, like you mention the caddies that catch a lot of folks out.  Buying a few thousand $ server is no big deal to me, but then I do that and get stuck with expensive mods ...and then I get pissed.  $400 and another $500 for some mods to get that snappy server is fine.  And I learn a lot.  I paid like $3000 in total for my Synology setup, so to split that into a SAN and a cool server I can do all kinds of other stuff with makes sense.  I can even sell the Syn and probably break even.

I treat this more like score in a game than real money, maybe like golf where you try and go low?  I try and get everything I want by imposing a budget, its good mental exercise.  If I fuck it up I would buy a second server or whatever to remedy it, but the point of this game is to get a kickass server for as little as possible.

[mapleLC]

One of my own firewall projects uses a MikroTik RouterBoard M33G, a dual-core MIPS MT7621A running at 880 MHz with 256 MiB of RAM, using OpenWRT (derivative).  It does also run some custom IOT and network connectivity monitoring stuff, plus Dynamic DNS and OpenVPN.  Cost is about 60€ new, plus an LTE modem unless you have wired Internet connection.

These things are the opposite of typical "servers" –– I like to use the term "appliance" –– exactly because they're so memory-constrained, even though they do do server stuff, and run a server-oriented Linux distribution.  For "reasons", mine does actually have a PCIe SSD (M33G supports up to 2242 size-wise), but typically there are also heavy limitations on storage use, affecting things like semi-permanent logging.

(I do also have more typical routers running OpenWRT, mostly acting as a bridge between 2.4 GHz and 5 GHz WiFi, and my wired untrusted LAN sections.)

Currently under development is a funky little service running on OpenWRT that consumes minimal RAM, and reacts to connection attempts by banning the entire IP address for a specified duration.  That is, you can still poke at the firewall ports, but any packets from that IP address on the internet will not be forwarded to the LAN by the firewall at all.  Access attempt pattern (ports and frequency) will determine the duration.  It's an idea I developed based on years of fail2ban use on non-memory-constrained machines.  I'm looking at wiring kernel netfilter packet logs also, so that direct (non-OpenVPN) connection attempts to specific ports on any LAN address will trigger that also, without having to be promiscuous.

I dont get it man.  Ignore, not ignore?  Or learn to take a joke.  We're all engineers. Friends, even.

I have a MikroTik, I have been learning about Docker and the idea of shifting my torrenting directly into a virtualized environment on my router and visible via a browser from anywhere in my LAN, and my router has a wireguard vpn bound to that docker instance, is pretty intriguing.

I am stuck getting the router to narrow the VPN to just 1 router port and anything downstream is automatically in the VPN.  It just doesn't work... yet.

I must say that getting up to speed with MikroTik is a pain in the ass, but now somewhat adapted, its a way better system than what I had with Ubiquiti, RouterOS is quite usable.

[mapleLC]

When I last worked with this stuff we were starting to see more and more blade servers, an entire computer on a cartridge that plugged into a rack mounted cage. It was all about higher density, packing more and more physical servers into a rack. They had little storage though, typically just a pair of 2.5" HDDs as I recall. These days everything I deal with is VMs, I don't even know what the actual hardware it runs on is, my current employer relies heavily on AWS.

I don't know the era you are talking about but its likely the same era where virtualization became the rage, in like 2 years we went from begging network admins for remote desktop access into servers to never seeing another base OS again.

The virtualization move only made sense from the perspective of cloud because when I saw it happen, I always led me to wonder why they didn't just design an OS from the beginning that kept this kind of containment. Oh. Its not profitable. AWS and Azure want it easy to virtualize, not contain an OS.  Virtualization is a way to sell subscription based services, the engineering problem is another one altogether, and quite a lot simpler.

[Halcyon]

In terms of quirks, all of them will have them to a degree, but with a lack of deep knowledge, I am looking to the community for some guidance on what to look for.

I would say, you have two options:

1. Build a PC yourself out of off-the-shelf hardware that has the specifications you desire.

2. Buy some second-hand servers (I tend to stick with Dell or IBM/Lenovo) with the specs you desire and experiment/tinker. But expect to run into unexpected issues if you are unfamiliar with enterprise hardware.

You mentioned "throughput" and multiple users, which is quite a broad statement. Do you mean throughput in terms of networking performance (i.e.: 10 Gigabits and higher), or disk performance, or are you "crunching numbers" and require high spec CPUs? It really all comes down to what the purpose of it is and what operating system and applications you're planning to run on it. There is no "best" server/computer. Budget is also a consideration and probably the most limiting factor.

When I last worked with this stuff we were starting to see more and more blade servers, an entire computer on a cartridge that plugged into a rack mounted cage. It was all about higher density, packing more and more physical servers into a rack. They had little storage though, typically just a pair of 2.5" HDDs as I recall. These days everything I deal with is VMs, I don't even know what the actual hardware it runs on is, my current employer relies heavily on AWS.

I don't know the era you are talking about but its likely the same era where virtualization became the rage, in like 2 years we went from begging network admins for remote desktop access into servers to never seeing another base OS again.

The virtualization move only made sense from the perspective of cloud because when I saw it happen, I always led me to wonder why they didn't just design an OS from the beginning that kept this kind of containment. Oh. Its not profitable. AWS and Azure want it easy to virtualize, not contain an OS.  Virtualization is a way to sell subscription based services, the engineering problem is another one altogether, and quite a lot simpler.

I configured virtualisation for a company I worked for around 2002/2003. I doubt you'll want hardware from back then?

[MK14]

I don't know the era you are talking about but its likely the same era where virtualization became the rage, in like 2 years we went from begging network admins for remote desktop access into servers to never seeing another base OS again.

The virtualization move only made sense from the perspective of cloud because when I saw it happen, I always led me to wonder why they didn't just design an OS from the beginning that kept this kind of containment. Oh. Its not profitable. AWS and Azure want it easy to virtualize, not contain an OS.  Virtualization is a way to sell subscription based services, the engineering problem is another one altogether, and quite a lot simpler.

The great thing about virtualization (once its learning curve has been sorted out).  Is that instead of perhaps needing 6 or 7 servers, but only buying (making do with) 4.  You can buy perhaps 2 good servers, one as a SAN/NAS and the other to run VMs on (full VMs).  You can then spin up as many VMs (within reason), that you need, on the virtualization server.

You can even mess around and experiment, to your hearts content.  With various VMs, creating them, shutting them down or even deleting them.  Until things are working out as intended.

wonder why they didn't just design an OS from the beginning that kept this kind of containment.

In a sense, they have that now.  By using Containers/Jails/Docker etc, it is somewhat similar, to what you just said.  I.e. Containers/Jails/Docker, let you do just that and effectively partition the same OS, into different containment sections.  Hence the terms such as Containers/Jails/Docker and so on.

If you start out with TrueNAS (for use as your SAN, as in another of your threads), on a reasonably capable server.  It should let you do virtualization and its version of Containers/Jails/Docker, as well.  It is NOT brilliant at that task (real dedicated virtualization use servers computers, is better), but good enough for initially playing around and later full-time light use.

EDIT:  Edited, to better differentiate, between full VMs (with their own OS, like a separate/new server or computer) and Containers/Jails/Docker, which are much lighter, resource wise.

[MK14]

In the light of a better idea on what you need, this suggestion could easily change.  But as a first throw, at something 'good enough' and real cheap, to start off from.

https://www.ebay.com/itm/175584212280?hash=item28e1a27538:g:cusAAOSwhN5jzq6A&amdata=enc%3AAQAHAAABIIbO8iMZnnPFtXirOTLr%2FdUkocPhxFc86ABAILCpV02MFBWxRCDhw5tqzVhuJIwfFkgHXJ%2BRl8PsczcWXgI2%2BRkMVgMhOM3xSeGf3Rj6HDjJ0eJY3MiQAFebxHfOxV2buQ%2B6jEg%2BGZp%2BMg4ryUjoXMf%2FeojElXHANMfj5z6%2BYD1qD5CrzR%2FeTyR4mAfukKrBbfP8f%2BQ3QRlzM9SWtDNuTm7UfdBzUwi25qB9EdL6F7e5GMlJ7SbCCZGTVWj0%2FACfwdZ8f3qfeKsfFfRSOXzZxJlB1QAuh3wAfCSPCIvb%2F84xqa0eLVP0QU7pSXcntmsLtVKXmyR5FZQiFjeZNW6Hem87hXC3DRJyjMs4eM3E034wABEiGZ6mjeXcU6kisogd9Q%3D%3D%7Ctkp%3ABFBMtO_1tdNh

Spec:
Lenovo P410 ThinkStation Intel E5-2680 V4 2.40GHz 14C 64GB 512GB SSD Win10Pro

Which is $273.

So, it is a Lenovo, which is one of the better makes.
In a tower case, which can be more convenient for home use, and makes it more like a PC.
14 Core, 28 Thread CPU (Broadwell), which is somewhat recent (opinions can vary).  14 is a fair number, and could make a number of VMs, at quite decent performance (especially for the money).
64 GB ram, which is plenty for many things, including various virtualization options.
The windows can be overwritten with Linux, as necessary.
A 512GB SSD, is a starting point, which could be expanded, in the future.

It seems to be missing a graphics card.  But you can either use the cheapest, basic display card (which is suitable/compatible) you can get (or already have), or choose something, better, depending on what you want to use it for.

At that price.  You would probably get much of your money back, when/if you have finished with it.

It would use more energy than a standard PC, but you are getting a big 64 GB of ram, and 14 cores, 28 threads, as opposed to perhaps 4 cores, 8 GB ram, if it was a used, older PC.

[DimitriP]

The E5-2680 v4 @2.4 GHz was dog slow in 2016 when it was "new". Slower than a CPU from 2014
Now it's an even slower dog!

https://www.cpubenchmark.net/compare/2779vs2275vs5060/Intel-Xeon-E5-2680-v4-vs-Intel-i7-4790K-vs-Intel-i7-13700K

Decommisioned servers are big, noisy, heavy slow and dirt cheap.
Hell, there are some new servers that are dog slow and not dirt cheap.
But that's another thread altogether.

[JohanH]

non-critical stuff like backup services

there are two kinds of people: those who don't yet consider backup services critical and those who already do.

My wording was wrong. I meant more like redundancy for (non-critical) services. With today's VMs you can quickly move services where you want them and it's still useful to have some extra hardware where you can have things running temporarily. Backup is entirely another class of things.

[Bicurico]

Hello,

A few years ago, I was looking at second hand servers on eBay, as well. I was surprised how cheap some were and quickly decided I want to have one!

I ended up purchasing three (!) IBM X3650 M3 rack servers over the period of 6 months. They were that cheap (100-200 Euro each).

What did I use them for? To be honest, the single use I gave them was to educate myself on the following topics:

- Windows Server
- Virtualizing Machines - using VMWare vSphere
- Having hosts for multiple different virtual machines
- Creation of parallel virtual network for IPTV applications (without crashing my regular home network)
- etc.

Considering this learning experience, the money I invested was exceptionally well spent. Because I am a teacher, too, I have access to educational licenses of Microsoft and Autodesk software, which were used for my testes and learning.

Using such servers as opossed to regular computers has a few advantages:

- You get remote admin modules on BIOS level, so you can remote control the server without having to use attached keyboard, monitor or mouse
- You get redundant PSU, in case something goes wrong (not very likely)
- You get LOADS of RAM - purchase wisely and the RAM already comes fitted in the machine, otherwise the RAM modules were cheap themselfes, so I have the servers with >64GB RAM (some 96GB, some only 64GB). This allows to test CAD/CAM projects without hitting the normal memory limit of i.e. my work laptop, which already has less common 32GB of RAM
- You get two Xeon class CPU's, which will be less performant than newer generation Intel Core i7/i9, but you will get more threads: on my servers I get 24 threads, as opposed to the regula 8-12 threads on an Intel Core i7. Note that you need to be doing parallel processing to take advantage of this.
- You can fit a lot of disks. One of the 3 servers came with 12 bays and yes, you can use regular 2.5" SDD or SATA drives

The cons are:

- The servers make a lot of noise, due to the PSU's and CPU coolers. A lot of noise means that I cannot leave the servers switched on at night, because I will hear them when in bed, which is a different floor!
- The power consumtion is noticeable, even when the servers are in standby (so that I can access the remote monitor tool that allows i.e. to switch them on). Due to current energy price, I have switched the servers off (unplugged the power cable)
- The CPU performance is less than a modern Intel Core i7 or Core i9. Modern Intel Core i9 come wih 16 threads (if I am not mistaken), so even that is no longer a big advantage anymore. I am only speaking of Intel, but go to AMD Ryzen and you get a single CPU with more threads than my servers.
- Spare parts can be expensive - it might be cheaper do actually buy a second server for parts! This is why I got my third server, but then I ended up to get all three working as desired...

Because of the success I had for MY GOALS, I ended up purchasing a second hand Dell server for the company I work for. This server cost about 1000 Euro and came with 4 CPU's and a total of 80 threads (!) and 256GB of RAM (!). I used this server to develop a cloud based CAM managing application that could significantly reduce CAM calculation time by splitting up one project in several parallel projects. We never sold this solution, but it was enough to show a big customer how calculation speed could be reduced - they stopped bugging us. The same application allowed us to show the manufacturer of the CAM software how they could increase calculation speed. This single application paid off the investment on this server.

Finally I ended up buying a HP Z600 workstation for myself, which uses the same processor and memory type than my 3 servers. With this I ended up with a workstation running on 96GB RAM and 24 threads. Impressive for CAD/CAM applications and cost me less than 300 Euro.

Why are used servers so cheap? Because they have been written off and because the performance / running costs are worse than newer generation servers.

Conclusion: for a running system, buy a new machine, as it will be cheaper over its intended life span (considering the energy price to run it).

If you want to own a rack server, because you want to learn and investigate how to use, what to use it for and because you find it cool: go for it.

Regards,
Vitor

[Bicurico]

Forgot to answer OP's question:

In my research I found that the most intersting rack servers are made by either IBM, HP or DELL.

I did not find great offers for HP machines, though.

You will find that different generations of IBM or DELL servers have their very own price range.

Considering your budget, look for what models fit the budget and go from there.

As a tip: do not try to buy a "better" machine outside your budget, because it comes with no memory modules, CPU's or disks, thinking you can buy them cheaply afterwards. That will be a losing game, as separate CPU's memory and HDD add considerably to the price.

The better purchase is to buy a LOADED machine that comes with lots of RAM and best CPU's of the family.

Also, make sure you get HDD bays! Servers often only come with plastic covers over the slots. To insert a HDD you need the bay. A bay can cost ridiculously more money than the price difference to an offer that includes them.

Finally, note that while a server can support a given CPU family (from worst to best CPU), you might need matching coolers! If you get the server fitted with the worst CPU and then purchase second hand CPU of the best kind, they might not work with included heat sink and cooler. Changing them might end up expensive!

Without specifying your budget and intended application, it is impossible to give you better advise.

Regards,
Vitor

[mapleLC]

In terms of quirks, all of them will have them to a degree, but with a lack of deep knowledge, I am looking to the community for some guidance on what to look for.

I would say, you have two options:

1. Build a PC yourself out of off-the-shelf hardware that has the specifications you desire.

2. Buy some second-hand servers (I tend to stick with Dell or IBM/Lenovo) with the specs you desire and experiment/tinker. But expect to run into unexpected issues if you are unfamiliar with enterprise hardware.

You mentioned "throughput" and multiple users, which is quite a broad statement. Do you mean throughput in terms of networking performance (i.e.: 10 Gigabits and higher), or disk performance, or are you "crunching numbers" and require high spec CPUs? It really all comes down to what the purpose of it is and what operating system and applications you're planning to run on it. There is no "best" server/computer. Budget is also a consideration and probably the most limiting factor.

Throughput is a generic word for the machines ability to receive, process, and deliver work, typically packets.  We used the word a lot when there were databases involved and you had to tune everything.

The word represents a holistic view of the machine's capacity for work, and throughput was a general measure of how well a system could perform as a unit, then again benchmarked when merged into the wider system.

Its back to purpose and the point of the search... I want a generic machine with a large capacity for work, I can tune it from there.  It seems one decision that is obvious is to get a machine already outfitted with 2 processors.  That hunt seems problematic.

[mapleLC]

In a sense, they have that now.  By using Containers/Jails/Docker etc, it is somewhat similar, to what you just said.  I.e. Containers/Jails/Docker, let you do just that and effectively partition the same OS, into different containment sections.  Hence the terms such as Containers/Jails/Docker and so on.

If you start out with TrueNAS (for use as your SAN, as in another of your threads), on a reasonably capable server.  It should let you do virtualization and its version of Containers/Jails/Docker, as well.  It is NOT brilliant at that task (real dedicated virtualization use servers computers, is better), but good enough for initially playing around and later full-time light use.

EDIT:  Edited, to better differentiate, between full VMs (with their own OS, like a separate/new server or computer) and Containers/Jails/Docker, which are much lighter, resource wise.

The Docker seems like a tool to fill the gap.  The same base operating systems are still there, perhaps save for Linux pared down to the basics to run Docker.  What I think makes sense here is to see an OS designed from the ground up, fit for purpose.  Virtualization has taken off to the point that its a need that someone out there could fill.  Who knows, someone is doing it perhaps.  But I don't mean taking Windows Server and dumbing it down so it uses less memory.  An OS designed for virtualization would be more of a terminal, as simplistic as possible with usability and stability being the core feature of the OS.  A virtual host OS can never crash, and windows crashes.

Its not topical, but has virtualization been done over POE yet?

[mapleLC]

Forgot to answer OP's question:

In my research I found that the most intersting rack servers are made by either IBM, HP or DELL.

I did not find great offers for HP machines, though.

You will find that different generations of IBM or DELL servers have their very own price range.

Considering your budget, look for what models fit the budget and go from there.

As a tip: do not try to buy a "better" machine outside your budget, because it comes with no memory modules, CPU's or disks, thinking you can buy them cheaply afterwards. That will be a losing game, as separate CPU's memory and HDD add considerably to the price.

The better purchase is to buy a LOADED machine that comes with lots of RAM and best CPU's of the family.

Also, make sure you get HDD bays! Servers often only come with plastic covers over the slots. To insert a HDD you need the bay. A bay can cost ridiculously more money than the price difference to an offer that includes them.

Finally, note that while a server can support a given CPU family (from worst to best CPU), you might need matching coolers! If you get the server fitted with the worst CPU and then purchase second hand CPU of the best kind, they might not work with included heat sink and cooler. Changing them might end up expensive!

Without specifying your budget and intended application, it is impossible to give you better advise.

I would agree that those 3 represent the larger market share and yes I have seen machines stripped of parts and it tends to cost more to finish outfitting it.

The only problem with those 3 brands is how proprietary they tend to be. I was wondering about Fujitsu because they are specifically not a software company.  Wouldn't that lend itself to the Fujitsu units being more generic in this sense?

The problem is they seem more expensive, I wonder why.  Is it because they are a smaller player, or are they making the better server with better internals.  Do you or anyone know the Fujitsus?

[M0HZH]

...
Its not topical, but has virtualization been done over POE yet?

Not sure what you mean. People are running RPis in clusters for years now, with only an Ethernet cable that provides power & connectivity and a managed switch to turn on/off/restart individual nodes.

[JohanH]

The Docker seems like a tool to fill the gap.  The same base operating systems are still there, perhaps save for Linux pared down to the basics to run Docker.  What I think makes sense here is to see an OS designed from the ground up, fit for purpose.  Virtualization has taken off to the point that its a need that someone out there could fill.  Who knows, someone is doing it perhaps.  But I don't mean taking Windows Server and dumbing it down so it uses less memory.  An OS designed for virtualization would be more of a terminal, as simplistic as possible with usability and stability being the core feature of the OS.  A virtual host OS can never crash, and windows crashes.

I wouldn't say never crash. We have three environments. VMware ESXi, Redhat Virtualization (RHV), which is KVM-based and Microsoft HyperV. The first two are pretty stable, the third one is the one that locks up now and then, maybe because it runs on Windows Server. ESXi is a barebones linux (you can even run it from an SD card) and RHV pretty basic Redhat. You manage both through a web interface. The successor to RHV is Openshift (but I haven't used it). VMware in my experience is good for mixed OS environments, whereas HyperV is used mostly for Windows and RHV for Linux (you can mix OS's, though). But this is my experience as a co-admin from a smaller environment of about 10 hosts and 30-50 VMs over a number of years. On workstation computers I've used Virtualbox and KVM/QEMU. There are pro's and cons of every virtualization environment.

[Bicurico]

"The only problem with those 3 brands is how proprietary they tend to be."

Hardware-wise, appart from memory modules, CPU's and harddisks, the rest is pretty proprietary. You won't be able to easily swap coolers, heatsinks or other components. Plus, most is soldered anyway.

Software-wise, these machines have specific applications for which they were designed. Because I don't have much interest in Linux nowadays, I can only reply the experience with Windows Server and VMWare ESXi: as the universe of rack servers is much smaller than generic PC's, both operating systems recognized the server right away - I did not have to install any drivers.

I guess this is the case for any generic mainstream rack server.

I don't know a thing about Fujitsu-Siemens server and i would expect them to be rather a niche product. That alone will raise the price: imagine you have a Fujitsu server running, which is doing its job for years. The saying "never touch a running system applies". What if suddenly a component of the server breaks, like the fan, the PSU or a component on the mainboard? Well, if I was the admin, I would buy a second hand server and exchange the disks or the broken component and would have everything running in no time. The sellers know this and adjust the asking price correspondingly.

Not the case with X3650M3 servers, as there are too many being sold.

I guess this explains the price difference.

[mapleLC]

The Docker seems like a tool to fill the gap.  The same base operating systems are still there, perhaps save for Linux pared down to the basics to run Docker.  What I think makes sense here is to see an OS designed from the ground up, fit for purpose.  Virtualization has taken off to the point that its a need that someone out there could fill.  Who knows, someone is doing it perhaps.  But I don't mean taking Windows Server and dumbing it down so it uses less memory.  An OS designed for virtualization would be more of a terminal, as simplistic as possible with usability and stability being the core feature of the OS.  A virtual host OS can never crash, and windows crashes.

I wouldn't say never crash. We have three environments. VMware ESXi, Redhat Virtualization (RHV), which is KVM-based and Microsoft HyperV. The first two are pretty stable, the third one is the one that locks up now and then, maybe because it runs on Windows Server. ESXi is a barebones linux (you can even run it from an SD card) and RHV pretty basic Redhat. You manage both through a web interface. The successor to RHV is Openshift (but I haven't used it). VMware in my experience is good for mixed OS environments, whereas HyperV is used mostly for Windows and RHV for Linux (you can mix OS's, though). But this is my experience as a co-admin from a smaller environment of about 10 hosts and 30-50 VMs over a number of years. On workstation computers I've used Virtualbox and KVM/QEMU. There are pro's and cons of every virtualization environment.

Without too deep a dive into virtualization of which my knowledge breaks down, but there are quite specific mechanisms that virtualization hits really hard, and my best understanding is we've pieced together enough with tools on hand to deal with it, but its pretty resource intensive as a result (generally speaking).

In fact, what I am probably saying is virtualization requires a redesign from the processor thru to the operating system.  Again limited, but my understanding is there are fundamental limitations in typical setups.  You sort of need a new SGI type company, but trying to get anything done now, at scale, is at least a 10 figure problem, then add to that training a whole ecosystem.  Linus might be the last of his kind mainly because what he did was possible when he did it.  I would imagine the kernel necessary for something like this would be quite incredible.

[magic]

And I'm the one trolling...

Indeed so.

No.

Trolling is the specific act of deliberately posting nonsense to get people triggered.
I'm triggered and literally shaking when people conflate trolling with plain stupidity.

edit
Maybe you are trolling me?
If so, good job, I fell for it

[brucehoult]

And I'm the one trolling...

Indeed so.

No.

Trolling is the specific act of deliberately posting nonsense to get people triggered.
I'm triggered and literally shaking when people conflate trolling with plain stupidity.

I admit your theory has merit.

But, how to tell the difference?

[shapirus]

But, how to tell the difference?

There was a good reason for the "not sure if trolling or just stupid" saying to come to life.

[MK14]

Unless I've seriously misunderstood things, and NOT trying to disappoint the OP.

There seem to be many, generic or very generic questions and no real idea of any kind of application(s), requirements, specifications, timescales, Hobby/University/Business, budget (saying really, really cheap but can go very, very expensive, doesn't really clarify the situation, that much), physical location of equipment, relevant equipment already owned and available, general background as to what is going on, etc.

Analogy:
I want an amplifier, cheap!

Would not really be enough on a forum like this, to help that much.
Audio/rf/specialist, output power, budget range, hobby/business, etc etc, to narrow it down.
But E.g. Guitar Amp, up to 50 watts RMS, already got speakers, max cost $200, for my hobby.  Would help a lot.

[mapleLC]

Unless I've seriously misunderstood things, and NOT trying to disappoint the OP.

There seem to be many, generic or very generic questions and no real idea of any kind of application(s), requirements, specifications, timescales, Hobby/University/Business, budget (saying really, really cheap but can go very, very expensive, doesn't really clarify the situation, that much), physical location of equipment, relevant equipment already owned and available, general background as to what is going on, etc.

Analogy:
I want an amplifier, cheap!

Would not really be enough on a forum like this, to help that much.
Audio/rf/specialist, output power, budget range, hobby/business, etc etc, to narrow it down.
But E.g. Guitar Amp, up to 50 watts RMS, already got speakers, max cost $200, for my hobby.  Would help a lot.

Its less misunderstanding and seems more wilful.  I provided the car analogy, you have your amp.  The wilful part is sidestepping the information I have provided and, like a few others, remain stuck ramming this notion of specificity before assistance. Read thru the thread looking for what could be crafted into requirements, there are probably 20 from what I have said so far.

The same forum was able to help me decide on a SAN.  Same "computers" subforum.  Strange. 

Overall: Generic machine with lots of capacity for work, more specifically...
Lots of processing power
Lots of throughput
Modern connection ports, etc
Dual procs
Larger form factor to accommodate change
Intention to work with a SAN
Linux
Emphasis on non-proprietary products
Emphasis on stability and quality
"Snappy" UI experience
Decommissioned price point (hundreds not thousands)
Fairly complete unit, no empty shells

I'm afraid I don't understand why this is not enough info because it seems like it is, so something more pedantic is happening, and now we see some people's lack of class is showing.