3 ms·

They are not.

dathinab 3y ago
it's complicated, but in general server centers can profite enormously from economy at scale, but also handling that many customers at scale adds additional cost compared to just having that much compute for your own use case, so does handing out tones of free credits to startups, and running a company (personal wise) at scale is often highly inefficient, too. Especial if you make things unnecessary complicated. but mainly, AWS and similar do not have a need to lower their raw compute price to that level because even without they get enough customers, e.g. there had been cases of Microsoft Azure in some locations running out of spare compute resources leading to issues for customer.
- goodpoint 3y ago
  > in general server centers can profite enormously from economy at scale That's plain false.
  - dathinab 3y ago
    no it's not most things compute related once you reach a certain minimal scale (i.e. need room cooling in your server room, etc.) get cheaper with scale sometimes quite massively to list some - hardware cost, sometimes you can't even get the hardware you need/want on smaller scale - electricity cost (through it's complicated, if you can't choose the place for the data center electricity cost only up to some location specific scale) - overhead for reliability (e.g. multiple power connections from different providers, multiple wired fast internet connections from different providers etc.) - fast low latency internet access (what apartments get is often not cutting it at all even for quite small scale) - man power for hardware maintenance (a single person can do hardware maintenance for quite a lot of units, having a team of such people makes handling sick cases, quitting and similar much easier) - custom software tooling (amortizes more) - physical access control (has a pretty high fix cost if you want to get it right, but after which it scales somewhat linearish, assuming we limit ourself to a single building or small complex) - revenue also get's more regulare with more customers, which makes expensive unepxpected revenue drops due to e.g. a customer going insolvent much easyer to handle and as such cheaper - hardware spare part handling gets easier and cheaper on scale - ... And sure there is an upper limit per-data center for scaling which is dependent on locality, but there are enough locations where this limit is quite high. And sure if you run an such a small scale that you don't even have to consider having a "proper" server room with cooling system which is more then just fans moving heat out of racks you often also can cut many other costs, like power redundancy, network redundancy, 24/7 hardware fixing support, etc. Through then if you have some B2B services with reliability guarantees you often can't cut that costs no matter how small the scale (that btw. can be a reason enough to use the cloud, providing the necessary reliability with custom hardware on a small (compute) scale is very expensive as it involves many high fix costs) Idk. if there was a misunderstanding so just to be clear, this is about cost of a data center per-compute unit getting cheaper with scale (up to some limit). Not about running your own ad-hoc reliability nightmare datacenter vs. renting cloud resources.

They are not.

it's complicated, but in general server centers can profite enormously from economy at scale, but also handling that many customers at scale adds additional cost compared to just having that much compute for your own use case, so does handing out tones of free credits to startups, and running a company (personal wise) at scale is often highly inefficient, too. Especial if you make things unnecessary complicated. but mainly, AWS and similar do not have a need to lower their raw compute price to that level because even without they get enough customers, e.g. there had been cases of Microsoft Azure in some locations running out of spare compute resources leading to issues for customer.

goodpoint 3y ago
> in general server centers can profite enormously from economy at scale That's plain false.
- dathinab 3y ago
  no it's not most things compute related once you reach a certain minimal scale (i.e. need room cooling in your server room, etc.) get cheaper with scale sometimes quite massively to list some - hardware cost, sometimes you can't even get the hardware you need/want on smaller scale - electricity cost (through it's complicated, if you can't choose the place for the data center electricity cost only up to some location specific scale) - overhead for reliability (e.g. multiple power connections from different providers, multiple wired fast internet connections from different providers etc.) - fast low latency internet access (what apartments get is often not cutting it at all even for quite small scale) - man power for hardware maintenance (a single person can do hardware maintenance for quite a lot of units, having a team of such people makes handling sick cases, quitting and similar much easier) - custom software tooling (amortizes more) - physical access control (has a pretty high fix cost if you want to get it right, but after which it scales somewhat linearish, assuming we limit ourself to a single building or small complex) - revenue also get's more regulare with more customers, which makes expensive unepxpected revenue drops due to e.g. a customer going insolvent much easyer to handle and as such cheaper - hardware spare part handling gets easier and cheaper on scale - ... And sure there is an upper limit per-data center for scaling which is dependent on locality, but there are enough locations where this limit is quite high. And sure if you run an such a small scale that you don't even have to consider having a "proper" server room with cooling system which is more then just fans moving heat out of racks you often also can cut many other costs, like power redundancy, network redundancy, 24/7 hardware fixing support, etc. Through then if you have some B2B services with reliability guarantees you often can't cut that costs no matter how small the scale (that btw. can be a reason enough to use the cloud, providing the necessary reliability with custom hardware on a small (compute) scale is very expensive as it involves many high fix costs) Idk. if there was a misunderstanding so just to be clear, this is about cost of a data center per-compute unit getting cheaper with scale (up to some limit). Not about running your own ad-hoc reliability nightmare datacenter vs. renting cloud resources.