Re: [gaia] [Stackevo-discuss] 5G: It's the Network, Stupid

[Joe Touch <[e-mail]>]

On 12/22/2015 7:48 AM, Theodore V Faber wrote:
> On 12/21/15, 15:55, "Stackevo-discuss on behalf of Joe Touch"
> <[e-mail] on behalf of [e-mail]> wrote:
> 
>>
>>
>> On 12/17/2015 9:49 AM, Linda Dunbar wrote:
>>> I strongly support the concept of network slicing for Applications or
>>> IoT networks. 
>>
>> FWIW, I do not - in specific, I support the notion of per-service
>> overlays, but would not call them "slices".
>>
>> Slices are an artifact of an OS-view of the network. It's a network
>> partitioning model that considers cross-overlay interaction only as a
>> violation of the model itself.
> 
> Just to defend my tribe a bit, I’d say more of a telco-view of the
> network.  Providing isolated strictly defined services - e.g. 3kHz voice
> channels - lets you reason about partitioning and service development both
> at the cost of maintaining a stranglehold on what gets deployed.  Slices
> only enforce their isolation and guarantees if all the equipment has
> capabilities and is management appropriate to it.
> 
> Yeah, there’s an OS/distributed systems feel to that, but because the
> constraints now extend out of a chassis and beyond a LAN, telco analogies
> make more sense to me.
> 
> I know you (Joe) know that, but I think it’s worth saying to the group
> because the more strictly you adhere to a slicing model the more
> constraints you put on your equipment and management. ...

That's true, but a separate issue of deployment feasibility.

Whether an overlay/virtual net has service guarantees or not, it's quite
different to have a model that ends in network addresses at interfaces
(overlay/VN) vs. ending in processes (typical of slice models).

When a network ends in a process, cross-network communication (gateways,
relays, application transits, or whatever) always become counterexamples
to the model rather than a natural part of it.

Joe