Thinking about #financial stability in the #Fediverse, which isn't an issue at the moment, but wondering if there are ways we can lower #costs of #hosting...
-
Thinking about #financial stability in the #Fediverse, which isn't an issue at the moment, but wondering if there are ways we can lower #costs of #hosting...
-
Raccoon at TechHub :mastodon:replied to Raccoon at TechHub :mastodon: last edited by
For instance, could servers with lots of interaction share some of their backend? Like if all the Tech servers used the same image hosting so we weren't downloading multiple copies of the same images?
-
Josh Riversreplied to Raccoon at TechHub :mastodon: last edited by
@Raccoon interesting…masto-netlify for content. Reliable, deduplicated storage for backups, Reducing computer costs by sharing propagation and subscriptions would likely be a nightmare of distributed systems problems. It would be fun to think about what application changes would be needed to support independent, but federated hosting/compute with an eye towards reducing duplicated costs (or letting individual hosts run with less headroom), that could turn into both community-run and SaaS cost improvement options.
I wonder if that “high road” would be easier/more cost-effective than mundane performance and stack optimization.
-
Raccoon at TechHub :mastodon:replied to Josh Rivers last edited by
@joshrivers
I'm not sure this would be that hard actually. Systems exist within Fediverse software for various different backends for image storage, and it's entirely possible to use an image storage system that accounts for that, or to just bypass that entirely by setting it up such that image data is retrieved directly if it's from a specific image server rather than trying to download it. That's the real problem that we're looking at here, that images are hosted on a bunch of different servers, when we could be clustering them, reducing the cost for groups of large heavily connected servers by massively reducing the amount of data they have to store individually.This might be worth looking into.
-
Josh Riversreplied to Raccoon at TechHub :mastodon: last edited by
@Raccoon Do you envision it as shared-with-high-trust where all the writers can be believed to be storing the correct content? Or would there need to be a mechanism for ensuring that the image written was the one posted by the OP? If trusted, then I’d think it would be pretty easy and you’d only need remediation for occasional bad-first-download failures. But to scale more (and save more) I’d think you’d eventually need to protect against every image being watermarked with an ad by spammers. (Authoritative writes might be answer, but then you’d need every publishing server to join in….
(All of the above is completely naive ideation)
-
Raccoon at TechHub :mastodon:replied to Josh Rivers last edited by
@joshrivers
I think I was pretty clear in my first post: this would be between servers with high interaction levels and close bonds. With TechHub, it would be like, Infosec.Exchange, Hachyderm, Tech.LGBT, DAIR, furry.engineer... These are instances we have cross traffic with, and while I don't always agree with their moderation teams, I absolutely trust them to be good custodians of a shared image server.