Re: ttys

Tim Newsham <newsham@lava.net> writes:

>...
>This is getting somewhat off topic here, but I dislike the heavy
>use of absolute port names for similar reasons. I understand
>their utility in situations such as bootstrapping, but going
>through the mount table provides greater flexibility. If
>a utility or server opens net/inet:tcp/clone
>then there is no way to override the networking stack (short
>of providing a cmd line argument to override the path). If on the
>other hand the path /net/tcp/clone is used, then the utility
>will happily work with whatever service is mounted there, wether
>it be the local networking stack, some sort of socks proxy, or
>a remotely mounted networking stack. Plan9's 8.5 uses this
>mechanism to great effect.

One early reason for this was the impact per-user mount tables (design
requirement for me) implemented in the user's state underneath the C library
(this *is* a microkernel) had on scalability. For quite a stretch of VSTa's
time, there was a round-trip exchange at each fork() for each connected
server. This has (happily) gone away, thus it's a lot less painful to put
more stuff in the mount space.

Andy
Received on 29 Jul 2004 17:53:02 GMT