Is pluggable TCP/IP stacks feasible in mainstream operating systems?

[LoadWB]

On Amiga we have been graced with AmiTCP, Termite TCP, Miami, Genesis, and probably other TCP/IP stacks about which I do not know.  IIRC, these mutated from an original stack produced by Commodore (AS225?) and all offer some compatibility to what appears to be ubiquitous among Amiga, the bsdsocket.library.

So I read about how Gibson Research decried the raw socket access introduced by the new Windows XP TCP/IP implentation (which has not caused the end of the world, best as I can tell,) and Windows Vista introduces another TCP/IP stack.  All of these harken back to winsock.dll and winsock2.dll.

Then there's the TCP/IP stack within the Linux kernel, and found in most Unix implementations such as Solaris (/dev/tcp, /dev/udp, etc.)

We run into so many issues with vendors' TCP/IP stacks (like Windows XP SP2's half-open connection limitation,) why do third party vendors not create third-party TCP/IP stacks?  Or do they?

Regardless of the thought process behind the curiosity, could we speculate on the viability?  Would it be a potential segregation of the mainstream OS world, or could one vendor's better implementation take over?

I see potential for the server market where many system builders, administrators, and maintainers would like to tweak system performance and security as much as possible.  Would TCP/IP outside of the operating system allow for such an approach?  And would it be too much of a potential black-eye for OS vendors to ever allow?

[Matt_H]

Hmmm... Interesting points. I'm far from being an expert on networking protocols, but I get the impression that (users of) other OSes don't perceive networking functionality as a 'stack', so to speak - TCP/IP is so tightly integrated into other systems that I don't think it's recognized as the independent component that we see it as.

The Amiga is interesting in the fact that our core Internet applications were built during an era of zero first-party system development. Things probably would have been different without the loss of Commodore. Envoy (which I have become a HUGE advocate of :-)) was supposedly due to be included with OS3.1. It's not a TCP/IP implementation, but it can be extended to be so, and I suspect that it eventually would have been. Maybe Kickstart 3.4 would have included envoy.library in ROM. Then we'd have had the same integrated networking as other systems. If that was the case, would developers be sufficiently inspired to write their own stacks? Would they be able to bypass the built-in stuff to do so without breaking compatibility?

In other words, do users/developers on other systems recognize TCP/IP as something they can improve upon, or do they just write it off as an OS-level flaw and hope it'll be different in the next version or on a different system?

Hopefully that made sense... ;-)