This post got over 50 comments a couple yrs ago:
https://linkage.ds8.zone/post/367822
Looking through the comments shows a strong bias in favor of centralization. Is it fair to say the masses in the fedi are merely anti-GAFAM? Considering how people pile onto a Cloudflare node (lemmy.world), the fedi is indeed mostly composed of those who do not resist network effect. Choosing the most evil node of a less evil network still reflects a problem.
We need to evolve more. We need a platform that does more to thwart centralized power. The network needs to build in a penalty for disproportionately large nodes w/a single point of influence. And there should be a built-in mitgating mechanism to keep Cloudflare out, as Cloudflare is inherently centralized to the most abusive extent.
Everyone has their reasons for defending centralization. For you it is cost savings, apparently. Money wins, unless the tech is done well.
(edit) Small instances are cheaper to secure because they are less prone to targeting. The bigger an instance gets, the more it costs per user for infosec.
It would be good, but not for the bottom line. Facebook has mastered the bottom line indeed.
It’s a problem for sure. Lemmy devs are pro-data portability, but they give it a low priority. They hope that someone will build a bot that ports data so long as it does not require Lemmy server code to be written. IOW, they (Nutomic specifically) calls for a hack job. He posts pseudocode here: https://lemmy.ml/post/31128657/19022698
The point in more users leading to a healthy revenue wouldn't be to support a profit-seeking company—something that many users are specifically here to avoid—but rather to ensure the instance is cushioned against unforeseen events and cost increases. Instances that can be expected to last beyond the short-term to medium-term should be backed by a non-profit organization with commitments to users to remove key man risk.
Even under the scenario of full data portability, switching instances creates confusion (particularly for casual users who just want to create an account and not worry about whether it will still exist in a year), eliminates the archive of content on the shuttered instance that will no longer federate, and breaks @ mention links, so safeguards to reduce the risk of instance failure outweigh the downsides of reduced decentralization.
Given that the software is being developed by literal communists and isn't in the hands of the Lemmy.world operators, I don't think limited centralization is likely to lead to user base exploitation or a particularly strong network effect.
100% availability would be nice, but there are trade-offs. Concentrated power on LW is a bad trade. Defeats the problem activitypub was meant to address.
Sure, stability is good, so long as the instance does not sell souls to Cloudflare or grow disproportionately big.
Data portability is not at odds with decentralization. It’s the other way around. A good degree of data portability promotes decentralization precisely because disaster can be mitigated when small intances fall.
Or are you calling centralization the “safeguard”? What a horrible trade to make. The right answer is better software, not centralization.
Look at the numbers. Network effect is in full force. Lemmy.world (Cloudflare centralized) is huge, an order of magnitude bigger than the 2nd biggest (sh.itjust.works IIRC, also Cloudflare). Dev.lemmy.ml was the biggest originally, and also Cloudflared to enable unhealthy growth. The devs wisdom improved and they scaled back deliberately. Now lemmy.ml is no longer centralized on Cloudflare (which was quite an embarrassment for founders of an activitypub implementation) and much smaller than LW (though still more than 2 std deviations bigger than the mean).
If an LW admin does not like what you write, you lose reach to a large portion of the fedi because of 1 person’s influence. Fuck convenience; esp. when convenience comes with Reddit-like power abuses.
w/a iow
what are you doing with all the time you’re saving?
I'm not sure I agree that infosec costs would scale with number of users. Your tech stack is either vulnerable or it's not. A larger server would mean you're more of a desirable target, but a small server (which "ideally" every server would be) would still have those same vulnerabilities.
And servers need to be cost effective. Not everyone can afford to run a server out of the good of their heart, and those that are able to do that are already preselected to be fairly wealthy. So yes, even in a decentralized system money will always win.
Two servers can have the same vulns but it’s not just a vuln-counting exercise. They don’t have the same threat model. There is no need to control for a vuln that will not be exploited, or where exploitation has negligable effect. A small instance that supports a dozen users running client X need not control for a vuln that only impacts users of client Y (to give a simplified example).
Small servers have more options for controlling for vulns. They don’t have to worry about a countermeasure breaking some obscure client as they don’t have a broadly diverse number of software configs connecting. By contrast, a large node is limited in how it can control for vulns due to having to support many different client configs from a diverse range of networks.
If there are thousands of users, then all possible varieties of client configurations are in play. And also the stakes are also higher (1000s of outraged ppl). Along with higher stakes is also the fact that there is a much greater chance of a sophisticated user getting pissed off at the administration and attacking. LW likely has countless inside adversaries.
There is even a greater chance of accidents being exploited on large nodes. E.g. the lemmy bug where a random user arbitrarily got admin privs spontaneously due to a glitch when the server was simultaneously accessed by many users. These sort of bugs don’t manifest on a node of a dozen users.
Small nodes can trivially stay current on versions. Whereas large nodes need to plan their migrations. The window of lag time gives an opportunity for attack.
A residential broadband connection and a cheap throwaway PC will do. A small instance accumulates ~30gb/year. I suppose it depends on one’s own perspective whether that calls for “wealth” but where I am PCs are dumped on curbs that can handle the workload.