[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: source address rewriting and shim6 proxies




El 29/09/2005, a las 20:28, Erik Nordmark escribió:

marcelo bagnulo braun wrote:
what difficulties am i missing?

The first one is that we don't know what problem we are trying to solve with shim proxies.

:-)
as you stated it in the previous message, there are at least two flavors of proxies: one would not require shim support form the the host (support legacy hosts) and the other one would require some support from the hosts (i.e. it would require hosts to support part of the shim)

one reason for the first type of proxies would be to enable shim functionalities in legacy hosts
The reasons that i can see for the second type of proxies would be that:
- it would offload the exit path selection to the routing system, which is more similar to the current paradigm. In this case, the exit path selection process would benefit from routing information available in the routing system of the multihomed site. In particular, exit path failures would be directly detected by the routing system. Moreover, it would enable the performance of some type of traffic engineering based in the the configuration of the route selection in the routing system, again more similar to current paradigm. - it would provide ingress filtering compatibility, leaving source address dependent routing just as a mechanism to support legacy hosts (internal and external). so even if there is still a need for something like source address dependent routing, the upgraded nodes won't need it.

It could be enabling legacy IPv6 hosts behind a proxy (this is what Paul is suggesting as far as I understand), or it could be having shim aware hosts offload some functionality to a shim proxy.

The issues with the two are quite different. So which one are we trying to solve?

Your comments were related to the second problem:

I am not sure i can see any difficulty with this... I mean in current multihomed site scenario, where the prefixes are quite stable and they change because of renumbering events associated to changes in the ISPs, the coordination of the prefix set perceived by the host and the prefix used by the routers seems feasible to me... In more dynamic scenarios, this may be require further analysis.

But this might be problematic if you have shim6 hosts with different capabilities. For instance, a smallish host is today allowed to not configure IPv6 addresses for all the prefixes announced on its links, which it might do to save on memory (think small battery operated devices).

Having the routers be configured to rewrite to any of the site's prefixes would then mean that the peer would see packets with a source locator which was not setup as part of the context.

right
as i see it, this capability should be optional. I mean, hosts should be able to prevent rewriting (this probably needs a don't rewrite bit as well :-( So, small devices as you mention may just prefer to do all their shim processing themselves.

Even if you want to ignore small (battery operated) IPv6 devices, you have similar issues when the set of prefixes change; the border routers notion of the prefixes to use will be out of synch with what any particular host in the site knows as the prefixes.


ok, as is see it the problem occurs when there is a new prefix available in the exit router that the hosts is not aware of, right? In this case, the router would stuff a prefix that is not in the locator set for the shim session and packets would be discarded.

I think that this may be an issue depending on the considered scenario.
For instance, if you are considering a traditional site i.e. a SOHO or big organization network, i guess this would be acceptable. I mean, in this case, the prefix changes when there is a renumbering event (because of changes in the ISP prefixes) This is a managed event and the addition and removal of prefixes is a managed event. So i guess that easily we could assume that the new prefix will be first notified to the hosts and then configured in the exit routers.

However, in more unstable scenarios, like those related to mobility, this may be an issue. I mean, if we are considering a mobile network that changes the prefixes very dinamically, then maybe it is not possible to assume that the hosts will be aware of the prefix before that the router use it (in particular, because in a mobility scenario, the new prefix may be the only available prefix) However, i am not sure if this would be much of an issue because in this case, we have other tools to deal with this (such as the nemo protocol)

One way to handle this is to have the receiver, instead of looking up the context based on <source locator, destination locator, context tag>, only use <destination locator, context tag> for the lookup. That way if the border routers put in a source locator that the source host isn't aware of, the packets still get accepted by the destination.


but this would mean that you would be accepting packets with any source address? wouldn't be some serious security issues there?


regards, marcelo

But as I said, I don't think this is the problem that folks are trying to solve with the external shim.

   Erik