Re: [gaia] What if all phones were always on the Internet?

Very interesting thoughts Dirk.

Another keyword is transport performance (related to the transport encryption discussion): with heterogeneous access, and especially with low-bitrate, unreliable secondary access, you want to have a more powerful forwarding plane that can react to changing link layer characteristics locally and that can employ in-network storage for reacting to disruption/delay. Today, mobile network operators are trying to address some of these issues with TCP proxies and application-layer proxies (the latter will probably become more expensive in the future because of encryption, see below).

A pointer on this is our recent work on In-Network Resource Pooling (in ACM HotNets 2014): http://www.ee.ucl.ac.uk/~uceeips/files/inrpp-hotnets14-ipsaras.pdf

According to this framework, in-network storage is utilised to temporarily store/cache data *mid-path* when some link along the path is broken or very slow/congested. Then transmission resumes when conditions get back to normal. It sounds like a good fit for low-bitrate, unreliable links.

Thanks,
Yiannis.

On 27 Nov 2015, at 08:51, Dirk Kutscher <Dirk.Kutscher@neclab.eu<mailto:Dirk.Kutscher@neclab.eu>> wrote:

How would that best be progressed? I think there are good ideas here (adding
delay-tolerant features to mainstream protocols) but I'm not sure how to get
those ingrained in e.g. 5g stuff.

In Europe, review and have your say on what's happening here: https://5g-ppp.eu/

There is a second phase of projects in preparation -- that should ideally cater for topics like this, but people need to speak up.

Technically, I think that ICN as a concept has good potential as a network infrastructure technology that can span both well-connected traditional type of networks as well as less conventional ones, including uni-directional and disruption-prone networks.

One of the keywords is access network heterogeneity, e.g., "hybrid access" in the current IP access world. It's obvious that TCP/IP (even MPTCP) with DNS/CDN has issues in even the simple hybrid access home gateway scenario.

Another keyword is transport performance (related to the transport encryption discussion): with heterogeneous access, and especially with low-bitrate, unreliable secondary access, you want to have a more powerful forwarding plane that can react to changing link layer characteristics locally and that can employ in-network storage for reacting to disruption/delay. Today, mobile network operators are trying to address some of these issues with TCP proxies and application-layer proxies (the latter will probably become more expensive in the future because of encryption, see below).

Finally, security: I see two things happening:

1) HTTP/2 and TLS => ubiquitous encryption
2) CDN extending into core and access networks => many vulnerable platforms that manage keys and certificates on behalf of original data owners and service providers to maintain connection-based encryption

This does not look very promising to me. We ought to figure object-based security and fine-granular access control based on that.

With virtualization extending to all parts of the network, it now becomes feasible to prototype and evaluate such systems, e.g., as a slice in a virtualized mobile network.

There is some interesting work going on, just a few pointers:

http://conferences2.sigcomm.org/acm-icn/2015/proceedings/p189-auge.pdf

http://conferences2.sigcomm.org/acm-icn/2015/proceedings/p137-schneider.pdf

http://conferences2.sigcomm.org/acm-icn/2015/proceedings/p177-yu.pdf

Cheers,
Dirk
_______________________________________________
gaia mailing list
gaia@irtf.org
https://www.irtf.org/mailman/listinfo/gaia