[AusNOG] AAB Statement

[Adrian Chadd]

On Wed, Sep 01, 2010, Jason Ashton wrote:

> >From page 3 - " End-user latency <10mS"
> 
> In my experience with next generation radio technologies the latency for large packets is typically double the latency for small packets, which again is hardly application impacting. And future generations of wireless systems will have even lower latency. 
> 
> We see latency as low as 25 microseconds (0.25 msec) on our high capacity point to point microwave links.

I've not gone digging into it and thus it's completely possible that something
in the 3G-nextgen spec(s) cover this, but just to be "correct":

Your equating of a PtP microwave link behaviour as somehow indicative of the
behaviour of a shared (time, code, frequency)-division media is a bit off-base.

I'm not talking about minimum latency, I'm referring to the maximum latency
on the last-hop. Obviously(!) the provider may have latency creep in due to
"the internet", so I'm focusing on last-hop.

Current 3G "internet" tech has a -wide- range in latency. Perhaps the spec
allows for IP-style "QoS" and a maximum per-user connection latency but this
isn't how it's deployed in the real world. From my very, very coarse
understanding, 3G low-latency is achievable only with significant drops in
client carrying capacity and link throughput. Again, I haven't done any
in-depth digging; I could be wrong. If so I'd appreciate a bit of schooling
in this matter.

Your PtP link has that low latency because it's just that - point to point;
single user or a small number of users. 802.11 also has similar issues -
pre-11n definitely has upper limits on subscriber count because of the
small (sub-few hundred) packets per second you can throw across the wire.
Add too many concurrent clients trying to do stuff and you end up exceeding
how many pps you can actually transmit. 11n changes the game a bit
(as there's packet aggregation magic in there which I haven't yet had
to implement; so I don't yet understand it fully) which changes the
behaviour profile from pre-n 802.11 behaviour.

(And strictly speaking - the same issues can and have crept up with non-
wireless tech; it's just the per-subscriber bandwidth nowdays doesn't
(totally) thrash the "medium" bandwidth limits - for example, the backplane
of your DSLAM/DSL aggregation device/dialup unit/etc. Some devices
implement that as a shared medium. :)

In short - "Comments not fully thought out; more data needed thx."



Adrian