At the
risk of over simplifying this.
1)
Deploying anything 4x faster than what you need is not cost-effective, ever.
Even deploying GE where 2xFE would work is more expensive.
2a) If
(again, thinking IXes here) you are offloading most of your locally sourced
traffic to peers at an IX, you may be able to use >OC48 connect
speeds
without needing your backbone to actually pass 20+Gb/s. Everyone has a
different network design, so it really depends. Guys who push can use 10GE
sooner (IMO) than guys that pull because of the IX case
here.
b) Cable networks and networks where most of the traffic is internal or to
a few large peers could benefit here too.
3a )
Anyone who doesn't have 5Gb/s of aggregate traffic probably doesn't have the
peer density to send more than 2Gb/s to a single IX or peer anyway. (see
#1).
b) In the case where at a single point you need more than 1-2Gb/s per peer, you
may want to deploy 10GE or something similar because you have sufficient
capacity to handle another peering location to fail entirely for an extended
period of time without (hopefully) affecting bandwidth to your peer. There are
some assumptions here, so YMMV.
Fortunately, no one is requiring anyone to use this,
yet...
Deepak
Jain
AiNET
The backbone at the time of my original work that I participated in was
40Gits/in and 40Gbits/out unless that has changed 10GigE is not practical or
cost effective if it is limited to local area's and provate connections. That
doesn't mean from A design
perspective that A cost effective solution has already been
designed, the position
of the market and the cost per megabit for most companies is not there,
most
companies now do 2.5Gbits bi-diectioonally for 5Gbits and barely use all
of that.
-Henry
Deepak Jain <deepak@ai.net>
wrote:
>
> While there are some smitherings about 10GigE, there are
>
technical reasons and
> > market reasons it is not really ready for
prime yet, that is
> not to say it's not going
> > to happen,
it is just not going happen now.
> >
>
> Some people
are using it in the MAN and WAN now though.
Exactly. At the EQIX/ASH
GPF Telia and AOL both said they were using 10GE
cross-connects for
private peering. So that means at least 3-4 major
networks are using them
in production in a LAN, MAN or WAN environment.
When you are
aggregating lots of a GEs, there isn't really a great,
cost-effective way
to move all of these bits cost-effectively. nxOC48 is
pretty cheap, but a
little ugly if you need the bandwidth unchoked. 10GE is
supposed to get
there, but at a 10xGE price, not a OC192 type price.
The real
advantage of Copper 10G is that eventually you can deploy it to all
the
existing copper [inside] plants that people have currently deployed.
Just
like GE, it eventually just becomes tolerant enough to use
existing
wiring. I would be very happy if the first boxes that came out
with these
long range xenpaks were muxes that would take 10xGE ->
1x10GE -- this would
solve the uplink problem from smaller gear in a
heartbeat.
Deepak
Jain
AiNET