I would say it depends on your definition though.
I used to rent 1U plus power and network in a rack on a single rented tile that was part of a rented room in a data center.
The owners of the data center rented out rooms, network, and power, and left the rest to tenants.
I see nothing wrong with such a data center. And they still exist; I know of at least one in the neighbourhood of the place that I moved out of a few months ago.
@farfalk
Oh you can totally do that if you want to and your home Internet connection can carry the load. I've done this myself and there's nothing wrong with it.
But there are cases where that isn't enough, or where a home Internet connection of sufficient speed is prohibitively expensive.
@cwebber @farfalk
That depends on where you are.
I read a blog post of someone once who got a 25Gbit line for his home Internet connection in Switzerland. I was like, why on earth would you do that... until I discovered that the price of that connection was about the same as the 100Mbit line I had at my place in South Africa at the time, with a 1Gbit line being more than double that price.
I wouldn't Colo in Switzerland. I might in SA.
@cwebber @farfalk
- replies
- 2
- announces
- 0
- likes
- 0
@cwebber @Colman @mirabilos @Reinald @thomasjwebb may I ask, how would you describe the change between then and now? I've never been in a DC and want to better understand when centralization turns noxious
@xlrobot
Let me try: I see 3 major facors:
Uniform scaling
Energy density
Cooling requirements
Colocation was wild - you rented Rackspace, and had the "remote hands" move your gear in. Everybody was using different makes. Storage servers, Compute Power, maybe some Backup Tape Library, A router. A full Rack had 6 to 8KW of power consumption.
Our webhosting DC had about a Dozen Racks in a room with wooden office doors, cooling was a monoblock AC.
@xlrobot
That was 2012ish.
We build a new location: Massive Storage. Gradually storage filled full racks. No longer a NetApp with 3 or four Disk extensions, but full racks. And Servers, i.e. compute, became blade systems, HP and Cisco. People got virtual machines instead of rackservers.
Energy consumption went to > 10KW in a rack.
In-Row active coolers, air-sealed rows. Cooling had about 50% footprint of the server racks.
@xlrobot
At the same time "high Performance computing" became a thing. Massive parallel compute clusters. Nations were competing for the top 500 List. We had a small HPC Cluster - for testing. Special network architecture, part of the nodes with nvidia cores for CUDA. The parallel computing was scientific. Protein folding, number crunching, engineering simulations.
Those facilities had usually uniform racks build and preconfigured.
@xlrobot
And management of those structures became automated: hundreds of nodes registered themselves on bootup, picking their jobs, starting to work. A single failed node was no reason for a change. 10% of degraded nodes were ok. Then you might swap a complete rack.
Even full containers in some facilities.
The Cloud DCs are similar. Uniform structures - automated management - large scale maintenance. No longer "remote hand admin" pushing a reset.
@xlrobot
And the server room envirement got hostile for humans: automated fire extinguishing systems flooding rooms with toxic gas - data is more valuable than human life. Humans are not supposed to work there any more.
For me the DC thing became "toxic", when they grew from basement facilities with a dozen racks to dedicated industrial facilities. Even though the human requirements require only a laughable small space.
@cwebber @Colman @mirabilos @thomasjwebb
@Reinald @xlrobot @cwebber @mirabilos @thomasjwebb they were dedicated facilities renting out rack space with exciting fire suppression from about 1996? 1998? That’s how telecoms centres were configured here anyway.
@Colman
My experiences are from University DC. That was a funny mixture of "lagging behind" and "bleeding edge".
Universities have been driving HPC, but business hosting (with continuity considerations) was not the main focus. Telecoms as infrastructure had definitely a different perspective.
@xlrobot
I.e. running a webshop, running SAP ERP-Software, all that requires just minimal space these days.
We are wasting 100 times more energy rendering an GenAI Marketing Video than it costs to sell and ship your product - that is where it has become toxic.
@Reinald @xlrobot @cwebber @mirabilos @thomasjwebb yeah, I guessed. The commercial stuff was like that before we hit critical mass. Smaller scale, certainly, but industrial facilities.
@Reinald @xlrobot @cwebber @mirabilos @thomasjwebb well, once they allowed stinky internet people into their nice machine rooms. Before that it was shelves or racks in basements, but that’s not a cost effective way of doing things when a leased isdn line has a three month lead time.
Ah, the good old days.
@Reinald @xlrobot @cwebber @mirabilos @thomasjwebb in my experience they haven’t been human centred *ever*. Anything that let me spend less time in there was a good thing.
@Colman
I recall visiting our Colocation Hoster in Munich 2010ish, that was a basement room in a regular office building. There was a Telecom Fiber Connection Hub next door, because connectivity was way more important than power supply. Short fiber to telecom was way cheaper than renting a leased line over kilometers, location mattered for connection.
Big HPC facilities were closer to power sources most time...