[aleph_minus_one]

> End users don't buy CPUs and buy filesystems. They buy entire systems. [...] Possibly true, but still totally irrelevant if the rest of the system can't keep up.

At least historically for hardware components of PCs, this was not irrelevant, but the state of things:

You basically bought some PC as a starting basis. Because of the high speed of improvements, everybody knew that you would soon replace parts as you deemed feasible. If some component was not suitable anymore, you swapped it (upgrade the PC). You bought a new PC if things got insanely outdated, and updating was not worth the money anymore. With this new PC, the cycle of updating components started back from the beginning.

[kstrauser]

But that still doesn't save away, "oh, it's only slow because the filesystem is so slow". Assuming that's true, that's a very integral part of the system that can't readily be swapped out by most people. You can't say "the system is actually really fast, it's just the OS that's slow", because the end result is just plain "the system is slow."

[aleph_minus_one]

> You can't say "the system is actually really fast, it's just the OS that's slow", because the end result is just plain "the system is slow."

If performance is so critical, people do find ways around this. Just to give an arbitrary example since you mention file systems:

Oracle implemented their own filesystem (ASM Cluster File System (ACFS)):

> https://en.wikipedia.org/w/index.php?title=Oracle_Cloud_File...

"ACFS provides direct I/O for Oracle database I/O workloads. ACFS implements indirect I/O however for general purpose files that typically perform small I/O for better response time."