Re: [uml-user] user-mode port 0.42-2.4.5

[Simon B. <si...@ci...>]

Hi Jeff

On Tue, 5 Jun 2001, Jeff Dike wrote:

> si...@ci... said:
> > implies that if you don't stick an ip number on the eth0=tap... call,
> > then the helper isn't needed - is that not the case?
> 
> Oops, no that's not the case.  The helper won't do any of the host setup, but 
> it is still needed as an async IO thread, since UML needs SIGIO for all its 
> inputs, /dev/tap* doesn't do SIGIO, so the helper is still needed to pass 
> packets back and forth.

Ahh, goodo.  I've got the CVS version of umn_net, and we're cooking with
gas!  Using pretty much the default setup from the networking.html file,
I've noticed one little oddity so far:

If I ping from the host box to the usermode box for five minutes or so, I
get this:

simon@gazza:~$ ping 192.168.0.254

--- 192.168.0.254 ping statistics ---
361 packets transmitted, 361 packets received, 0% packet loss
round-trip min/avg/max = 0.1/0.1/0.3 ms

All shiny.  Over exactly the same period of time, pinging in the other
direction, I get:

usermode:~# ping 192.168.0.251

--- 192.168.0.251 ping statistics ---
73 packets transmitted, 73 packets received, 0% packet loss
round-trip min/avg/max = 0.2/0.2/0.3 ms

There seems to be less pings sent in the same five minute(ish) period.  Is
this expected behaviour?  It looks kinda weird, watching ping stutter it's
way from ping to ping, but without actually dropping any
packets.  Interestingly, I noticed identical behaviour using the slip
based setup last night, so it's not specific to ethertap.

> > Sorry, what I was really after wasn't a .deb file, just the uml helper
> > binaries in the deb-package-2.4.5.tar.bz2 - it's got everything else
> > you need, but then you have to fossick about either in the .rpm or in
> > CVS to get the helper apps.
> 

> Are you saying that you weren't asking for a .deb then, or that you
> don't want a deb because it will do you no good?

Well, a .deb is certainly better than a .rpm, for those of us who use
Debian, but I kinda liked the tbz2 file, for understanding what was going
on.  However, for long term sync of kernel and userspace packages, with
sensible boot scripts, .deb would be useful.
 
> I'm trying to retire the packages.  They were OK as a getting-started
> package, they're no good for shipping userspace stuff like helpers and
> daemons.
 
Yep, I can see that.  It's pretty hard to enforce dependancies in a tar
file.  A .deb version of user-mode-linux-0.42_2.4.5-0.i386.rpm, and
seperate .deb versions of the indivudual root file systems would be a
pretty cool thing.

Cheers
Si