Re: [Softwires] A few draft-dhankins-softwire-tunnel-option-03 comments

[Sheng Jiang <shengjiang at huawei.com>]

From the title of this draft, it only serves the dual-stack lite scenario.
For 
support generic tunnel scenarios, we have recently proposed a new draft
draft-guo-softwire-sc-discovery
http://tools.ietf.org/html/draft-guo-softwire-sc-discovery-00 
to softwire WG. It proposes a general softwire (tunnel) concentrator 
discovery mechanism. It is generic to server different kinds of
tunnel/softwire
concentrator auto discover scenarios includes the scenario of 
draft-dhankins-softwire-tunnel-option.

Regards,

Sheng

> -----Original Message-----
> From: softwires-bounces at ietf.org 
> [mailto:softwires-bounces at ietf.org] On Behalf Of Tomasz Mrugalski
> Sent: Monday, June 29, 2009 7:55 AM
> To: softwires at ietf.org
> Subject: [Softwires] A few 
> draft-dhankins-softwire-tunnel-option-03 comments
> 
> 
> Hi,
> I've read draft-dhankins-softwire-tunnel-option-03 draft 
> recently and have several comments and questions:
> 
> 1. Tunnel type
> Can we assume that IPv4-over-IPv6 will be the only tunnel type used?
> I think not. Therefore tunnel-type field should be added to 
> the DS_LITE option. One octet is enough.
> 
> 2. SI hints
> I think it should be possible (or even recommended) that 
> client will send not only OPTION_DS_LITE in ORO, but also MAY 
> provide OPTION_DS_LITE itself in SOLICIT, REQUEST, RENEW, 
> etc. messages. It should contain address of previously used 
> SC address, provided as a hint to the DHCPv6 server. Here's 
> scenario, where this approach could be useful:
> 
> Network Operator may have more than one SC. It is reasonable 
> to assume that client-to-SC assignment will not be fixed, but 
> rather client will be assigned to a SC (e.g. due to load-balancing).
> 
> Using knowledge about previously used SC, Network Operator 
> may prefer to assign this particular client to the same SC as 
> before. There are several benefits of such assignment. In 
> particular, if SI was reboot, some enc-user sessions may 
> still be active.
> 
> 3. Preference
> In section 3, multihoming is mentioned. Would it be 
> beneficial to provide additional field in the option called 
> Tunnel preference? Its meaning would be similar to the one in 
> DHCPv6's PREFERENCE option. In case when multihoming device 
> has several softwire offers, it would use (or favor) the one 
> with highest preference. The best size for this field would 
> be 1 octet.
> 
> 4. Renewal
> According to Softwire Problem Statement (RFC4925), section 
> 1.1, softwires are generally dynamic. Shouldn't SC address be 
> periodically renewed? This would be useful for graceful 
> failover to a different SC. There are several possible solutions:
> a) Include current SC address in all RENEW messages, using 
> DS_LITE_OPTION.
>     (poor choice, renewal frequency depends on leased address and
>     prefix T1 values)
> b) use Information Refresh Time Option (RFC4242) to define how often
>     DS_LITE option should be renewed.
> c) add T1 (and possibly T2) to the DS_LITE option, with 
> meaning similar to
> T1 and T2 used in IA_NA and IA_PD options.
> 
> Another issue is related to the following question: Is 
> softwire permanent? 
> Can we assume that softwire is permanent in the sense that 
> once established, it will be used and is valid until SI 
> crashes or is powered off? What about failover scenarios when 
> SC malfunctions or reaches end of its lifecycle? Maybe it 
> would be useful to specify lifetime? This counter could be 
> reset every time DS_LITE option is renewed, exactly the same 
> as addresses in IA_NA option.
> 
> Size of those fields (T1,T2, lifetime) would be standard 
> double words and values expressed in seconds.
> 
> 5. Softwire termination
> Again, Softwire Problem Statement (RFC4925), Section 1.1 
> states that softwires may be initiated and terminated on 
> demand. How should SI release its tunnel? Assuming that 
> softwire tunnel is teared down once SI sends Release message 
> is not appropriate. I think the simplest approach is the
> best: SI should include OPTION_DS_LITE in the RELEASE 
> message. Server may acknowledge this by sending DS_LITE 
> option with address set to ::. This could be used to notify 
> SC that it may free its resources associated with tunnel for 
> this particular SI.
> 
> 6. Stateless autoconfiguration
> DS_LITE on its own, without address being delegated, is 
> useless. I think clarification should be added that this 
> option may not be used in stateless DHCPv6 (i.e. is not 
> allowed in INFORMATION-REQUEST message).
> 
> 7. Leasequery
> There's a mechanism for querying server about provided 
> configuration parameters (RFC5007). Would it be useful for a 
> server to report that DS_LITE option was provided to the 
> client in question? I'm not sure if such details should be 
> put in the draft or not.
> 
> I think that's it for now. That's my first post on IETF, so 
> forgive me if I was pointing out the obvious or discussing 
> topic already covered (I've checked list archives, but I 
> could have missed something).
> 
> Regards,
> -- 
> Tomasz Mrugalski                Dibbler - a portable DHCPv6 
> implementation
> tomasz.mrugalski at klub.com.pl    http://klub.com.pl/dhcpv6/
> Gdansk University of Technology
> _______________________________________________
> Softwires mailing list
> Softwires at ietf.org
> https://www.ietf.org/mailman/listinfo/softwires