Re: Mount groups

From: Andy Valencia From: <vandys_at_nospam.org>
Date: Sun Mar 07 2004 - 11:54:27 PST

>Imagine if something like gnome was started with the current vsta and
>you want to mount a cd. then around 10 apps would have to mount it
>individually.

Ah, OK, so you're thinking about dismountable media.

>you could just read it every time you need it. would require a few more
>context switches every time a file is opened though.

This would load the general, common, case with the cost of handling
the relatively rare case. Probably a thousand or more CPU cycles
overhead per file open.

One thing I will point out is that all servers are always available
from your filename space anyway. If you have a (newly started) server
at //disk/fd then anybody who wants to can do stuff like:

        dd if=//disk/fd:fd0 of=my_fd_img

The "//" construct reaches out to the namer on the fly, and gets
talking to the registered server. The part after the "//" until the
":" is the namer name, and everything after the ":" is processed
as if disk/fd had been mounted somewhere in the first place.

In fact, for dismountable media (in VSTa, the code to run the media
itself is dynamic) it might make sense to just spruce up the support
for symlinks so that we could do "ln -s //disk/fd:fd0 /dev/fd0" and
thus you would reach the server when it's running, and get ENOENT
when it wasn't. Yeah, UNIX would've returned ENODEV, but everybody
knows I take artistic license with POSIX. :-> And now you have only
a cost of some small files in the filesystem to carry.

Andy
Received on 7 Mar 2004 19:54:27 GMT

This archive was generated by hypermail 2.1.8 : Tue Sep 26 2006 - 09:03:09 PDT