Why Plan-9?

[Sunder]

On Sun, 21 Oct 2001, Jim Choate wrote:

> Transitive means that A mounts B, C mounts A and gets B free. Plan 9 does
> this, managed by a set of authorization layers for fine control, native.

This could be bad.  Say B doesn't want to allow access to C for its file
systems. Then what?  Any mechanisms to prevent A from resharing it?  It's
also bad depending on how it's implemented.

Say A has a low bandwith connection to B such as ISDN, but it has a high
bandwith connection to C.  If C uses A to get to B, that slows you
down.  Does it treat the file systems differently or is it just resharing
a mounted file system?


> This means that when Hangar 18 goes online you can mount /hangar18 into
> your filespace (via Plan 9 or Linux NFS services) and you will get all the
> resources that Hangar 18 mounts through that point. ftp is a good example.

Yeah, so now you're eating double the bandwith of hangar 18.  Say hangar
18 is connected over a T1 and nothing else.  You mount it's file systems
over nfs or whatever Plan 9 uses and do cd /ftp2.sourceforge.net, then you
copy a file from there.  

Well now, you've just effectively doubled the traffic to hangar18.  If the
owner is paying a flat rate, it's not so bad, you're just wastng bandwith.  
But if it's at a colocation it will be enough to go over the percentile
rating and have to pay for extra usage.

Even if the two boxes are on the same local network (broadcast domain),
you'd still saturate your hubs/switches by doubling this traffic, and even
slow down access (latency) as you're using plan-9 to translate from ftp to
it's native scheme to nfs.  Further, you're eating lots of extra cpu
cycles on the plan 9 box(es) while they do this. 

What if hangar18 has a few ftp sites mounted and you run the equivalent of
the find command (which recursively searches for files.)  It goes over the
ftp connections and walks every single connection of the internet?  That's
kind of insane.  It'll take forever and eat up bandwidth at both ends.



And as you say below, it would even cache some of it... ugh...  hope there
are limits to this.

> In Plan 9 you 'mount' the ftp server to your file system. If you ever go
> out and walk that part of the file space tree and request a file it only
> then goes and gets it. You can control its lifetime (to manage disk space
> for example) via local cache controls. A 'lazy update' mechanism, very
> efficient of network and local resources.

Yeah, it's "cool" but wasteful.

> The network layer. The traffic between any two Plan 9 boxes is encrypted
> with keys dependent upon the individual boxes (or larger classes if you
> desire) if the system is so configured. You can also use this to encrypt
> branches of your filesystem. Plan 9 provides SSH.

Just DES? And no disk encryption????  What about swap?

Any ip filtering / firewalling mechanisms?  Any routing mechanisms?  Any
NATting mechanisms?

The clustering / distributing computing aspects of it are
cool.  But... I'm still not convinced...
 
> > >  - Doesn't use passwords, Instead it uses tickets (ie certificates).
> > 
> > ...which are granted via...?
> 
> However the resource owner choses. I'm using 'small worlds network' models
> for my 'web of trust'...

So it's like kerberos in some ways?
 
> It should be, X Windows sucks.

Xwindows by itself without a window/desktop manager? yes that would suck
balls.  XWindows with Ximian GNOME or KDE, nope - it rocks, though not on
old hardware.

> > What's Inferno?
> 
> Another OS, intended for real-time control of "Internet Appliances". You
> run it along side Plan 9.

Whoop. :)  You can control most internet appliances through a web gui or
telnet/ssh command line anyway.