RE: DNS discovery

[<teemu.savolainen@nokia.com>]

>     teemu> I'd like to point out that any additional (round trip)
>     teemu> delay in configuration phase, especially in wireless
>     teemu> (cellular) environments, is _not_ user friendly. Please
>     teemu> consider those 200.000.000+ million cellular users opening
>     teemu> a network link for browsing etc. purposes who have to wait
>     teemu> every time for yet another round trip .....
> 
> While I agree that minimising delays and RTTs is always desirable,
> your argument is a bit of a non-sequitur. There's usually a delay of
> 10-15 seconds after a mobile phone is powered on or roams from its
> home provider before it has finished the authentication dance and
> joined some mobile operator's network. Given that context, an extra
> RTT to find a DNS server is no big deal in the overall scheme of
> things. In fact I suspect that the name server location overhead could
> readily be assimilated into the "authenticate this handset" overhead.
> Hey, perhaps getting a valid DNS server address could *be* 
> that handset
> authentication. 

Well, I'm not expert of all wireless networking technologies as I'm been working mainly with GSM (much less with CDMA/TDMA/PDC/WLAN etc.). Therefore I'm not sure if your argument is true in some networking technology. 

In current GSM phones no IP address allocation or any such configuration is done when powering on the phone and registering to a network, or when roaming to another operator's cellular network. 

IP address allocation (and other IP-related configuration like DNS discovery:-) is done when GSM cellular phone creates a connection to IP network - via GPRS PDP context opening or CSD dial-up. It is possible to have a cellular phone in GPRS always-online state, where this configuration is of course done only once and overhead is small as you described.

However, often (depending on use-case again) connections from (GSM) cellular phones are opened only when needed in order to save power, operator resources, money etc. Therefore a connection is opened every time a user wishes to browse, send MMS/email, use Java for networking and so worth and closed when that need ends.

So IP connections are opened many times more often than a GSM cellular phone registers to actual cellular network infrastructure (depending of course what user actually does..). Thus this RTT has to be waited every time such activity takes place and not only in powerup and when registering to a network.

I do not know often average cellular phone user would get this RTT penalty in real life, but I think that avoiding this RTT would be a good design anyhow.

What comes to DHCPv6 vs. RA, would it be possible for the hosts just to look if RA has any DNS options and use them if available, otherwise use DHCPv6(-lite) / alternative methods? That way those network administrators who wish to use RA could enable it to the routers and others could leave it out.

By the way, is there any documentation that discusses preference between preconfigured, DHCPv6 configured, RA configured, IPCP configured (and possible future IPv6CP configured), and well-known addresses?-)



Regards,

      Teemu


#----------------------------------------------------------------------
# To unsubscribe, send a message to <dnsop-request@cafax.se>.