Re: [dhcwg] WGLC on draft-ietf-pcp-dhcp-03

["Bernie Volz (volz)" <[e-mail]>]

Ted:

I think sending domains names for the client to process is far more dynamic. Also, the draft specifies use of domain names.

When does the DHCP server do this resolution (from domain name to address)?

-          During configuration (which doesn't allow for any change in the information) - i.e., when the operator first enters the information.

-          During startup (i.e., when the server reads the configuration) - (which would require a reload and might have serious performance issues if DNS is unavailable when the server is reloaded)

-          During each client request (which have serious performance issues if DNS is unavailable)
You can see that these three approaches (and perhaps there are more) are much less dynamic than having the client get the information when needed and also don't negatively impact the DHCP server (and thus other clients that might not be relying on DNS) when DNS is unavailable. One might limit the time one waits for the resolution, but what do you send if you can't resolve?

This is the grand old debate about whether one should send addresses, domain names, or both (in which case we've sometimes defined two options or defined suboptions or an encoding format). And, we sadly have never really solved this issue within the DHC WG.

The http://datatracker.ietf.org/doc/draft-ietf-dhc-option-guidelines does seem to indicate your solution, but it never specifies "when" this resolution should be done. I think that's likely and issue we need to discuss about this guidelines draft. (There's a few other oddities in that draft that should be corrected - why for example is it picking on DHCPv6 - "DHCPv6 stands for Dynamic Host Configuration Protocol for IPv6.  Contrary to its name, is many contexts it is not dynamic."; DHCPv4 was no better and actually worse since ForceRenew is really not implemented.) But that is likely for another email thread or for discussion on Monday at the DHC WG meeting.


-          Bernie

From: Ted Lemon [mailto:[e-mail]]
Sent: Friday, July 27, 2012 3:34 PM
To: Dave Thaler
Cc: Bernie Volz (volz); [e-mail]; [e-mail]
Subject: Re: [dhcwg] WGLC on draft-ietf-pcp-dhcp-03

On Jul 27, 2012, at 3:06 PM, Dave Thaler wrote:
Right.   I think the goal should be to minimize the amount of processing the
client has to do with the option contents... just pass it to getaddrinfo() or
equivalent was what the WG had discussed.

Remembering that the DHCP server automatically translates names to IP addresses in its configuration, why is it necessary to place this burden on the client?   Traditionally DHCP just sends IP addresses, not names, precisely because it minimizes the burden on the client while maintaining the flexibility of doing name translation on the server.   I assume you already considered this, because I'm pretty sure you've seen debates about this go by before, but I'm curious as to why you decided to send a name rather than an IP address.