[Ndn-interest] Issues in routing
Klaus Schneider
klaus at cs.arizona.edu
Mon Nov 19 13:33:00 PST 2018
> The nfd connection states that after installing NFD in a client machine,
>> we can specify the IP address and port number of the remote NFD, so that
>> NDN packets get encapsulated into UDP or TCP packets and sent to the
>> remote NFD. e.g. nfdc create udp4://192.0.2.1:6363
You are talking about running NDN over IP tunnels (here, UDP tunnels).
However, these tunnels only exist for point-to-point communication
between routers to make NDN run over IP networks. They could easily be
replaced by an Ethernet face (if there was a direct Layer 2 connection
between two routers).
When using the word connection in your earlier sense ("and the
connections [a router] holds", "take care the connections of the nodes")
I assumed you are talking about a transport layer connection as it
exists in TCP between client and server.
However, this transport layer connection does not exist in NDN. A router
does not maintain any connection state, thus nothing dramatic happens to
the packet flow after a router goes down. You will lose a number of
Interests/Data packets that would have gone through the router, but
these can quickly be retransmitted, and will find a different path to
the repository (if it exists).
> when a
>> new ndn node wants to enter in an existing network and establish a
>> connection with a node of the network, it does the above thing, isn't
>> it?
This is different from the idea of a transport-layer connection, as
described above. If your node could have a wired Ethernet connection to
an NDN router, this step wouldn't be necessary. You could just send
Interests to that router without needing an IP address.
Moreover, I don't think it's related to routers going down, unless you
are talking about your access router. In this case, you will (of course)
need to re-establish the tunnel to the access router after it comes back up.
Best regards,
Klaus
On 11/19/18 2:15 PM, Tanusree Chatterjee wrote:
> Hello Klaus,
>
> Thank you. It cleared my doubts to some extent. But, you said that there
> is no notion of connection in NDN and I did not understand what does it
> exactly mean.
> The nfd connection states that after installing NFD in a client machine,
> we can specify the IP address and port number of the remote NFD, so that
> NDN packets get encapsulated into UDP or TCP packets and sent to the
> remote NFD. e.g. nfdc create udp4://192.0.2.1:6363
> <http://192.0.2.1:6363/> states the client machine wants to establish a
> connection with 192.0.2.1:6363 <http://192.0.2.1:6363/>. Also as
> Junaxio's explained in my last email answer that NDN-FCH service tells
> us the IP of the routers near by. However, here I am a bit confused why
> there is no notion of a connection. As my previous question was when a
> new ndn node wants to enter in an existing network and establish a
> connection with a node of the network, it does the above thing, isn't
> it? Please tell me if I understood anything wrong.
>
> Thanks & Regards,
> Tanusree Chatterjee
>
> On Nov 19, 2018 11:08 PM, "Klaus Schneider" <klaus at cs.arizona.edu
> <mailto:klaus at cs.arizona.edu>> wrote:
>
> Hey Tanusree,
>
> I think it's useful to distinguish between 3 logical entities:
>
> 1. Router -- forwards packets towards a storage location
> 2. Content Store -- a temporary storage location (can be cleared at
> any time)
> 3. Content Repository -- a permanent storage location (in the same
> sense as today's web servers are "permanent")
>
> These can all be on the same machine or on a different machine. For
> example, a core NDN router will probably have a content store, but
> not a content repository.
>
> Now to answer your questions:
>
> So, if a router is down for some reason, what about the data it
>
> stores in its CS and the connections it hold?
>
>
> Well, the content store will be unreachable. There is usually no
> notion of a "connection" in NDN, so other routers should be able to
> fetch the data from somewhere else.
>
> If the router is
>
> temporarily down, it can resume to its normal operations
> sometime later
>
>
> Sure. Why not?
>
> But if the router is permanently down and there are several data
>
> it produce and there are no more copies of all the data.
>
>
> That only matters if the last content repository is down. In this
> case the data might become unavailable, if all the content stores
> have cleared the data as well (which they are free to do).
>
> NDN does not have the goal to permanently replicate every content
> piece inside the network. The content provider is still responsible
> for keeping their content repository up and running, and likely
> wants to provide some redundancy here.
>
> NDN, however, can help with the scalability: it reduces the load on
> the content repository when its content objects become more popular.
>
>
> Best regards,
> Klaus
>
>
>
> On 11/19/18 2:32 AM, Tanusree Chatterjee wrote:
>
> Hello all,
>
> In NDN the routers are the busiest and most responsible entity
> in the network. So, if a router is down for some reason, what
> about the data it stores in its CS and the connections it hold?
> If the router is temporarily down, it can resume to its normal
> operations sometime later. But if the router is permanently
> down and there are several data it produce and there are no more
> copies of all the data. Can there be any network administrator
> which can have the copies of the data of a router when it is
> down? If it is a high connectivity node, can network
> administrator can play a vital roll to take care the connections
> of the nodes?
>
> -- Thanks & Regards,
> Tanusree Chatterjee
>
>
> _______________________________________________
> Ndn-interest mailing list
> Ndn-interest at lists.cs.ucla.edu
> <mailto:Ndn-interest at lists.cs.ucla.edu>
> http://www.lists.cs.ucla.edu/mailman/listinfo/ndn-interest
> <http://www.lists.cs.ucla.edu/mailman/listinfo/ndn-interest>
>
More information about the Ndn-interest
mailing list