Re: Call for v6ops agenda items

[Cameron Byrne <cb.list6@gmail.com>]

On Wed, Feb 17, 2010 at 1:43 PM, Mikael Abrahamsson <swmike@swm.pp.se> wrote:
> On Wed, 17 Feb 2010, Cameron Byrne wrote:
>
>> Can you help me understand the scalability limits you are referencing?
>
> The MME offerings I've seen today seem to be in the 10M simultaneous bearer
> range. That means you'd need 400 MMEs to reach 4 billion bearers to reach
> the /32 allocation exhaustion.

MME are generally deployed on ATCA blades, so they are within scope of
Moore's Law on CPU.  MME interaction seem to really scale with
transactions for registration, paging, and handover ... So the call
model is relevant, and M2M as a major growth area is low MME
signaling.   I am not sure MME will be a major road block to realistic
scaling (4 billion not being realistic today).  If I have nearly 100
aggregation locations today, and I have a pool of 2+ MME in each
aggregation location to keep latency low and availability high,
200-300 MME  does not seam unreasonable.  Keep in mind i am talking
10+ years of planning, growth, ... So, not today.... I am just
theorizing about the limits of the system.  I don't disagree with you.

>
> I'd imagine the S/PGW are in the same scalability range, so I'd say that
> before you run into the /32 problem with IPv6, you'd hit a lot of other
> problems long before that.

That's why i am thinking S/PGW will be distributed to the cell site
location in the long term.  There is little value or sense in taking
LTE or LTE++  type traffic for a city the size of New York into one
location like we do today with 3G.  As a wireless only operator (not
the incumbent wireline carrier), it behooves us to exit Internet bound
traffic quickly.  That means at the cell site or some intermediate
aggregation location of   < 10 cell sites.

>
> Basically, the current LTE framework with GTPv2 tunnels for everything will
> run into scalability problems long before the current /64 per bearer
> framework will.

Agreed.  We cannot tunnel all traffic forever.  There  will have to be
mobility trade-offs made to scale the system.  EPC is not really that
different from the legacy packet core in that way.

>
> --
> Mikael Abrahamsson    email: swmike@swm.pp.se
>