DY and I are going private. If anyone wants a summary of this discussion DY 
will provide one.

Dino

On Jul 17, 2014, at 2:14 PM, DaeYoung KIM <[email protected]> wrote:

> On Thu, Jul 17, 2014 at 10:33 PM, Dino Farinacci <[email protected]> wrote:
> 
> > On Thu, Jul 17, 2014 at 10:03 PM, Dino Farinacci <[email protected]> 
> > wrote:
> > But when an EID is assigned to a node (either static or via DHCP)
> >
> > Is there a case in LISP that EID would be assigned to its interface rather 
> > than to a node?
> 
> That is independent of LISP.
> 
> Yes, in principle. But if you'd maintain the assertion (if such an assertion 
> has ever been made) that the network elements within a site doesn't change, 
> it would automatically assumed that
> 
>   - nodes in a site maintains the IP address (already) assigned, and
>   - the IP addresses would have been assigned to the interfaces.
> 
> And a node can have multiple EIDs if one chooses to do that. In fact, a 
> dual-stack host will have two EIDs, one IPv4 and one IPv6.
> 
> This multiple EIDs for a node is not my immediate concern in this discussion. 
> I'm try to confine myself to the most basic model of LISP, in order better to 
> understand it.
>  
> > Somewhere, it is said that the behavior of network elements within a site 
> > doesn't change, which is thought to be one of the major strong feature of 
> > LISP. If that is the case, then the EID would be assigned to the interface 
> > like with the current Internet.
> 
> Right, there are no requirements how EIDs are assigned locally within a 
> system. Remember, the EID is on a unchanged host. From that host's point of 
> view, it is just an address. From the LISP site's point of view, that address 
> is reachable via local routing.
> 
> This means... (if an EID is assigned to an interface)... the e2e transport 
> connection would break if a node move from one subnet to another within a 
> site. LISP doesn't contribute to solving/improving the intra-site mobility. 
> Right?
> 
> > Assuming you really meant the node, not the interface, which existing 
> > mechanism would allow the mobility? OSPF or ISIS?
> 
> What I meant is the "IP address" is moving. Saying it that way means no 
> matter how it is assigned on the local device, what is moving is the IP 
> address.
> 
> I'm still in the same hole. Does this IP address change when a node moves 
> from one subnet to another or not?
>  
> > If you meant, in fact, the interface but not the node in your discussion of 
> > the EID assignment, do you mean a MIP like mechanism would provide the 
> > mobility within a site?
> 
> I didn't and don't need to make that distinction.
> 
> So, you don't care about the intra-site problems at all. You focus on the 
> inter-site routing, don't you?
> 
> The whole idea of LIS(locator/id separation) is to solve, among many things, 
> the mobility problem. To ensure that, an ID would be kept constant at moving 
> of a node while locator would be updated at every such moving. Am I correct?
> 
> However, if behavior of the all network elements within a site doesn't have 
> to change, as asserted somewhere in the main documents, (this would mean that 
> ARP is also at work as usual), the ID (in fact, the IP address) would change 
> at any such moving, which would not keep the EID constant. Such move would 
> not only break a intra-site connection but also a connection between sites.
> 
> So, you cannot just ignore the intra-site problems. If the EID(IP address) 
> has to change at every node moving across subnets within a site, the whole 
> purpose of ID (in LIS) would be questioned. Isn't this in contradiction to 
> the whole idea of LIS?
> 
> If the EID would be kept constant even at moving within a site, then some 
> network elements within a site have to change their behavior, don't they? 
> Even the host behavior might have to change like, for example, shutting off 
> the ARP operation.
> 
> Also, EIDs can be flat, as with other usual LIS proposals, and need not 
> necessarily be out of a power-of-2 EID prefix. Correct?
> 
> > So in the static case, that would mean a host route would need to be 
> > injected into the IGP if the node's address stays the same
> >
> > I cannot catch what you exactly mean here by 'injection'. Sorry...
> 
> When a host 10.1.1.1 moves off subnet 10.1.0.0/16 to subnet 11.1.0.0/16, the 
> routers attached to 11.1 need to advertise a 10.1.1.1/32 route into the 
> routing system.
> 
> Got it. This is exactly what ISIS is doing, right?
> 
> > So, in short, what mobility solution is provided by LISP in addition to the 
> > existing mobility mechanisms?
> 
> There are different scopes of mobility. When you roam across LISP site, you 
> use LISP mobility.
> 
> You mean the tunneling..? Tunneling implicitly provides some sort of 
> mobility... like that?
>  
> When you roam within LISP sites you use IP-mobility or host routes. 
> 
> So, no additional contribution by LISP as far as intra-site mobility is 
> concerned. Right? 
> 
> -- 
> DY

_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp

Reply via email to