Re: [touch@isi.edu: draft may be of interest]

>I addressed how this affects p-HTTP in the implications section.
>The main problem is that muxing at the application layer has some
>side-effects, and will be incompatible with kernel-based integrated
>services scheduling mechanisms.

I don't really consider HTTP to be fixed at the application layer.
One of the things I have always wanted to do is to push the base
functionality of the web into the O/S. It should be possible to treat
URIs as filenames. IE open ("http://foobar.com/", O_RDONLY) or 
whatever your API happens to be.

NFS is in the kernel, at least as far as we should be concerned 
at the IETF. We are not USENIX after all.

>Granted, in-the-app is easier to deploy, but only because you're
>moving the kernel functions into the app, which can cause
>interferences later as kernel functions evolve. It also assumes
>that you're running only a single Web browser.

I'm somewhat pessimistic about Kernel functions evolving. Looking
at my UNIX box on my desk I don't see it providing any more features
today than it used to. Granted, more of those features actually work
rather than merely claiming to. I don't see a cross platform scheme
for file locking thats credible and threads seem to have only 
lukewarm support from the vendors.

What I would like to hear is how hard it is to modify Windows NT to 
add in a new protocol stack. At this point thats the only metric that
interests me because its the platform that has mindshare. If it is
possible to add a driver to support a new IP protocol to an installed
NT machine without upgrading the O/S then as far as I'm concerned new
IP protocols are a feasible route. If NT has a facility then UNIX
vendors that cannot duplicate it have a serious problem. I just got
quoted $800 for 64Mb of RAM. That indicates to me that the base level
PC can finaly run a real operating system which in turn makes me think
that most pcs will be shipping with a choice of NT 4.0 or Windows'95 
in a very short while.

To sumarise, I think we should forget about the travails of adding
protocols to a UNIX kernel, its the wrong mindset to be in. Forget about
designing systems for 1970s technology, the O/S scene has moved on,
wellcome to the 1980s :-)


		Phill

Received on Friday, 14 June 1996 10:33:59 UTC