If you mean the Inter Planetary File System, it uses DNS TXT records. But I don't know whether those are strictly required, or just a convenience. Ultimately, this all turns into an information theoretical problem: If one doesn't want to use the existing DNS, then one has to create something that is sort of similar, since an exhaustive search of the 128-bit IPv6 address space is impossible, and of the 32-bit IPv4 space impractical, in particular due to anti-hackery measures. If a system such as IPFS has its own way of locating host IP addresses, it's hard to imagine that the new way is significantly better than the DNS that we all have 40+ years of experience with.
Yes, systems that have never gone down do exist. They are very rare, and require a lot of work. Both at design and implementation, and during operation. My favorite example is from a slightly different field, namely the life support system software for the space shuttle, which was written by IBM Federal Systems (later a division of Loral, now part of LockMart). That software never had a bug. The cost was that programmer productivity was about 1/2 line of code per month.
Agree, there should be way more research (at universities and industrial research labs) and development (at both big companies and startups) on resilient systems. And on the related problem of self-describing data, which has mostly been looked at in the context of long-term preservation, not of avoiding dumb mistakes. Alas, the great big AI monster is currently eating all the available $ and brain cells.
All the cloud providers and many other (sometimes internal) systems today use distributed and decentralized databases. If you read their research papers, they typically have multiple layers of decentralized storage, building on each other, getting more and more resilient as you go down the stack. But the best theoretical resilience in the world doesn't help you if either (a) all eggs are in one basket, or (b) you rely on an external system to find the eggs in the great easter egg hunt. Funny example, again from a slightly different field: We all know how RAID works to make data tolerant to disk failures, by storing multiple (partial) copies of the data on separate disks (perhaps even in different locations or on multiple continents). Well, there have been bugs in RAID implementations that stored these multiple copies on ... drum roll ... a single disk ... ching boom! Not intentionally, I might add. Big oops. Cost me several nights of sleep.
