Crosspost: https://git.pleroma.social/pleroma/pleroma-meta/-/issues/43#note_67070
I've been reading into server costs, and starting to get a better idea about how this works.
Putting a server on the internet:
- Buy your own server hardware.
- Find a local datacenter.
- Rent rack space from them.
- You put your own server(s) onto the rack.
- Each slot on a rack is called "1U" (one unit?)
- 1U costs about $50-300/mo.
- You can buy servers that are 1U size, 2U size, etc.
- Your IP address comes from the datacenter.
This seems like a flexible enough system, I can start small and expand.
Let's say each Pleroma server needs 1GB of RAM, so I can support 144 customers on 144GB RAM.
I can buy a 1U, 144GB RAM server for $400: https://www.newegg.com/hp-proliant-dl360-g6-rack/p/2NS-0006-31892
I'm not accounting for disk space yet, but this seems more doable than I imagined. Of course I'm making a lot of assumptions. But even twice that cost still seems within reason.
I'd still need to figure out how to delegate new installations across multiple servers as I scale up. Someone pointed out the Elixir [Realbook](https://hexdocs.pm/realbook/Realbook.html) library earlier which seems promising.
I've been reading into server costs, and starting to get a better idea about how this works.
Putting a server on the internet:
- Buy your own server hardware.
- Find a local datacenter.
- Rent rack space from them.
- You put your own server(s) onto the rack.
- Each slot on a rack is called "1U" (one unit?)
- 1U costs about $50-300/mo.
- You can buy servers that are 1U size, 2U size, etc.
- Your IP address comes from the datacenter.
This seems like a flexible enough system, I can start small and expand.
Let's say each Pleroma server needs 1GB of RAM, so I can support 144 customers on 144GB RAM.
I can buy a 1U, 144GB RAM server for $400: https://www.newegg.com/hp-proliant-dl360-g6-rack/p/2NS-0006-31892
I'm not accounting for disk space yet, but this seems more doable than I imagined. Of course I'm making a lot of assumptions. But even twice that cost still seems within reason.
I'd still need to figure out how to delegate new installations across multiple servers as I scale up. Someone pointed out the Elixir [Realbook](https://hexdocs.pm/realbook/Realbook.html) library earlier which seems promising.
@alex Have you considered something like AWS EC2?
@valleyforge Definitely, way too expensive. I looked into it because it seemed Amazon would be less likely to deplatform arbitrarily, but it's prohibitively expensive. I was surprised.
@valleyforge Yeah but what are the specs on it, 0.5GB RAM? Even DigitalOcean's $5/mo plan has 1GB RAM.
@alex yeah my web server requires almost no ram
@valleyforge Yeah, in a VPS scenario Pleroma needs about 1GB RAM when accounting for the OS and runtime environment. I want to charge the end user $10/mo but on AWS I'd have to pay $10/mo. 😅 Self hosting is seeming more viable. But I wonder if they're seeing new customers right now, seems weird to be touring a datacenter during COVID but I guess it's fine.
@alex @valleyforge Don't forget about disk space, a typical pleroma instance will probably use more than 50GB of storage. Also, don't you have to pay for the bandwidth as well, on aws?
@r @valleyforge Good point about storage. Maybe we can deduplicate uploads between Tribes servers at least.
@alex @valleyforge It's mostly the database, on my instance, with hardly 2-3 active users, database size is 32 GB. You can remove old remote posts, but that breaks older threads.
@r @alex @valleyforge that seems pretty high. I think my instance is 8 gigs? Am I just not that federated or something?
