The internet? And many of its protocols that came with it, including email, etc.
Sidenote, and I dunno if it's what you're implying but many people do: "company x takes decentralized solution, offers services on top that aren't decentralized" doesn't make the whole thing centralized. Git is still a wonderful decentralized tool. There is very little functionality that Github offers on top of git that is particularly centralized -- permissions, maybe, but that's about it and it doesn't make much sense to centralize them. Nearly all of Github's featureset could be, like, a separate app that has nothing to do with git - it competes with the atlassian suite, for example; they just chose to make the projects "repo-centric" and make git a first-class citizen. They've also supported svn for a while.
GitHub's not just an index -- issues and pull requests are a core feature, and our usual decentralized tools for those fail here because you need write access to be able to place issues and pull requests into the main repo, yet you also want them to be creatable by anyone in the world while resisting spam and abuse.
Agreed with the idea that the internet is inherently decentralized. Too bad that we ultimately needed to succumb to some form of centralization in the form of ISPs in order for it to be a truly global technology. There are now some blockchain based hardware projects coming out trying to circumvent this.
And you're right about Github and git. I mean to say git couldn't really scale to the level of adoption it had without that centralized discovery layer on top of it.
Skype used to use each user's PC as a node, and every once in a while your PC would become a "supernode" and handle other people's traffic. The only part that was centralized was the login, as there was only one global namespace.  It was end-to-end encrypted too, tho it was proprietary and maybe backdoor'd but what's new. After microsoft bought it they started handling the supernode traffic. "nine months after Microsoft bought Skype, the NSA boasted that a new capability had tripled the amount of Skype video calls being collected through Prism" 
Yes I recall this. Too bad what MSFT did with it. I also noticed severe performance reductions when this eventually happened, but I wasn't entirely sure if it was due to a botched transition by MSFT or not. Needless to say, Skype has lost a ton of market share now.
It would have had to happen anyway. Skype's use of their customers' PCs for relaying traffic depended upon 1) users keeping their PCs turned on, and Skype running, constantly; and 2) users being tolerant of Skype consuming their computer's bandwidth and processor time. Both of these were starting to change by 2013, particularly as the market started shifting towards laptops (which are shut off frequently, often have less reliable wireless network connections, and make CPU consumption much more obvious by spinning up their fan).
Interesting lesson here I would say. It seems that relying too heavily on users with limited uptime computing power is not a good idea, until you reach a certain network size where it makes sense. The question now is, does this change with more computing devices with longer uptime (for e.g. a smart phone)?
I think it's important to distinguish between applications and protocols.
Successful applications based on decentralized networks seem to be a mix of both application code and protocol. So in discussing email, we are probably discussing SMTP and IMAP, and then the applications are the email clients that speak those protocols. Similarly, when we talk about the web we are talking about TCP/IP and HTTP(S), (also DNS, CGI, etc) and the browsers are again the clients that speak those protocols.
It seems to me that the success of the world wide web has entirely to do with TBL making sure that his superiors released his work into the public domain.
Whereas with something like World of Warcraft, there you have an extremely successful application, but it's based on a proprietary protocol, proprietary code, and centralized networks. Could it have been more successful if Blizzard had released the protocol and client into the public domain?
Freenet has been around since 2000, and I remember wondering why Wikipedia was always begging for millions of dollars when there was a way to have users host the content. Sure you don't want to ask someone to install a node just to view wikipedia, but there's enough editors and archivists out there that would be willing to be mirrors if it was just a background process you could run like freenet. It got me interested in programming and I learned that handling conflicts on a decentralized system makes things difficult, plus apparently there's more to running an encyclopedia than hosting servers.
Now wikipedia is mirror'd on IPFS, but its just a mirror, you can't edit. Still the performance of it is surprisingly good.
That’s a very sharp realization, that a lot of people don’t think about.
Email is the largest decentralized network in the world, with people routinely setting up federated instances. It’s the de facto standard protocol for b2b communication. It’s even used as a protocol beyond conversations, for meetings and notifications for example.
By far the most successful application of the ideals of decentralization.
Again, unfortunately for mass adoption to occur centralized solutions had to be developed to allow for wide market adoption, but yes I absolutely agree that Email IS the decentralized application. Too bad a lot of ISPs block outbound traffic on SMTP forcing you to use theirs or another centralized email solution.