[c-nsp] new 8-port 10 G bade

[Arnold Nipper]

On 29.09.2006 10:48 Gert Doering wrote

> Hi,
> 
> On Thu, Sep 28, 2006 at 08:31:20PM -0400, Richard A Steenbergen wrote:
>> On Thu, Sep 28, 2006 at 09:28:29AM -0400, Jared Mauch wrote:
>> > 	I think it's critically important for folks to continue to ask
>> > Cisco (and your other vendors) for a 100GE solution.  IEEE process is a
>> > bit odd, so could possibly stall.  If that's the case, us "nsp-types"
>> > may need to band together and have a SP nonstandard 100GE bakeoff.
>> 
>> At the risk of not being on the "we need more bandwidth now!!!" bandwagon, 
>> I don't understand why people think 100GE is so critically important, and 
>> why 40GE "just won't do at all". Parallel Nx10GE paths is a perfectly 
>> viable way to scale a network given the commodity technology currently 
>> available. Even given the 8 member limit that most vendors stop at 
>> currently, can you honestly tell me you have links where you need more 
>> than 80Gbps of capacity, and where a sensible architectre wouldn't call 
>> for adding a diverse path, another trunk, or another router anyways?
> 
> Look at some existing internet exchanges (AMS-IX, DECIX, LINX) - these are 
> hitting 30-40 Gbit inter-switch-traffic today, and growing 100 per cent 
> per year.
> 
> Given the nature of IXPs, there isn't very much you can do with 
> "Layer 3 balancing" - IXPs are L2 meshes, period.
> 

but you may interconnect those switches via L3 and tunnel L2 ...

> So you need either
> 
>   - 40G/100G channels (on Ethernet switches)
>   - more than 8 member links in 10G channels
>     (plus switches that permit you to actually connect 16x 10G to a box
>      without filling up most of the 10G ports you can have...)
>   - more intelligent L2 switch meshing than STP
> 
> having none of that is a problem for IXPs.
> 
> gert



Arnold, AN45