Hi,yuchi,

  PLS see inline.

Thanks,
Wei



"Yuchi Chen" <[email protected]> 2014-02-18 12:39:34:

> Hi Wei,
> 
> Thanks for the review. The purpose of this work is to provide a 
> unified and flexible framework 
> for implementing and deploying IPv6 transition mechanisms, and 
> enable these mechanisms 
> to coexist in the same infrastructure. Please see inline for the 
> detailed responses.
> 
> Best regards!
> --------------
> Yuchi Chen
> 
> On 2/17/14, 17:36 PM, "[email protected]" <[email protected]> 
wrote:
> 
> >Hi Yong, Yuchi, all
> > 
> >  This is a interesting topic. I reviewed this draft and  have some
> >comments and questions.
> 
> >  1) In section 5
> 
> >   "When receiving an incoming packet that doesn't have a match in the
> >   flow table (usually the intial packet of a new flow), CPE Switch and
> >   BR Switch MUST forward the packet to Controller.  Controller MUST
> >   determine how to proceed the flow (e.g. whether to apply NAT/ NAT64
> >   translation and/ or softwire encapsluaton/ decapuslutaion), and
> >   interpret the process into a set of forwarding rule configurations.
> >   Controller then passes these configurations to CPE Swtich and BR
> >   Switch.  CPE Switch and BR Switch then configure their flow table
> >   according to these configurations,"
> > 
> >  Why not let controller create flow table. Since packet is received by
> >controller, why not let it do more?
> [yuchi] In this framework switch should do packet forwarding, just 
> like traditional router (or L3 switch). The
> main difference is that switch doesn't know any other information 
> except for information of flows passing by, 
> which is maintained in the flow table. Controller - a centralized 
> (cluster of) device(s) at a higher level in the 
> network - knows everything (e.g. topology, routing, NAT states, etc)
> of the network. It can tell a switch how 
> to proceed and forward packets of each flow by configuring the flow 
> table maintained by switch. Hence 
> controller just needs to deal with the first packet of a flow, 
> determine how to proceed the flow, then let the 
> switch do the rest. I think it might not be efficient and may lead 
> to message congestion if let controller do all 
> the packet forwarding.
> 
> >  BTW, does "usually the intial packet of a new flow" mean an intial 
> >packet
> >of each session? If so, it might lead to message congestion between 
switch 
> >&
> >controller.
> [yuchi] Yes, switch should forward the inital packet of each of 
> unknown flows to controller. I agree that 
> it indeed may lead to congestion if there are too many new flows 
> concurrently passing through switch. 
> However, as controller at the same time only deals with packets no 
> more than the number of concurrent 
> flows passing through switches it connects to (note that all the 
> outbound and inbound packets of a session 
> can be regarded as belonging to the same flow), I think the 
> congestion is less likely to occur and can be 
> mitigated easily.

[Wei] I agree. Congestion might hardly occur in normal network traffic.
Anyway, the rate of  creating new sessions is a key performance indexes of
per apparatus under traditional routing scenario. 
  My proposal is that we should consider how to maintain existing 
performance
per apparatus (switch/router) while finding a way to unify IPv6 transition 

technologies.
  i.e. reasonable number of controllers in network...


> 
> > 
> >  2) What is the concrete consideration between CPE/BR & controller? 
> >such as netconfig, i2rs.
> > 
> [yuchi] The concrete interface (if this is what you mean) between 
> switch and controller is not specified. 
> IMHO all the protocols (or interfaces) that are able to carry flow 
> information and flow table configuration 
> can be adopted.
> 
> >  3) Logging is important in NAT, what is the consideration about it?
> >Maybe a switch is not suitable for logging itself.
> [yuchi] The logging of a switch can be done by its associated 
> controller. The controller is responsible for 
> maintaining all states that is related to the switch, such as NAT44 
> state and flow table (if needed). 

[Wei]If so, does a switch need to send its NAT44 sessions(or flow tables) 
to
the controller? 


> > 
> >
> >Thanks,
> >Wei
> >
> > _______________________________________________
> > Int-area mailing list
> > [email protected]
> > https://www.ietf.org/mailman/listinfo/int-area
_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area

Reply via email to