[f-nsp] Features on Brocade Ethernet platforms

Nick Hilliard nick at foobar.org
Sat Mar 12 14:16:02 EST 2011


On 12/03/2011 09:37, Robert Hass wrote:
> 2) TurboIron 24X - as core platform
>
> Is this switch is good solution to pure 10GE core ? We want use them for :
> - interconnect to other core nodes by 1x10GE and 2x10GE LAGs
> - interconnect to upstream providers (eg. GlobalCrossing, KPN and DE-CIX)

TI24X may be useful in this situation if you're aware of its limitations 
and how they might interact with your network.  It's a cut-thru switch with 
very small buffers (2M shared per chassis), which means that if you're 
microbursting *outbound* on your 10G ports, you have enough space for 1.6ms 
of buffered traffic, best case.  This may or may not work for your network 
configuration.

It also does L3.  I don't use this capability, so have no opinion on how 
good it is.

> - downlinks to PE routers by 2x10GE LAGs
> - downlinks to access switches (customer access switches)
> - at core we transporting multicast with IPTV (around 900Mbps) it will
> be not problem ? (eg. with microbursts at core)
> - is SFP+ ER (40KM) 1550nm is supported in this switch ?

As a general principle, I'm not convinced that running ER/ZR SFP+ optics is 
a good idea.  Apart from any heat dissipation issues (which may or may not 
be a problem on the TI24X, depending on the quantity of ER transceivers 
installed), SFP+ transceivers do not have on-board electronic dispersion 
compensation, but instead delegate this process to the switch motherboard. 
  This means that instead of having all analog->digital signals processing 
happening on the transceiver, with a purely digital electrical hand-off to 
the switch mobo (as you happen in an XFP), you end up with two potential 
kit manufacturers involved in the opto-electrical conversion process: the 
SFP+ manufacturer and the switch mobo manufacturer.

This may or may not matter to you, and it will depend entirely on the type 
of SFP+ used, whether it's been tested on the TI24X, and the dispersion 
characteristics of the fibre run.  However, if you're running on longer 
fibre links, I wouldn't do it directly into any SFP+ switch, not just the 
TI24X.  There's too much that can potentially go wrong.

If Brocade produce an ER SFP+ for this switch, you should be ok, but I 
don't think they do.

> - is QinQ supported on this switch ?

Yes, 4.2.00 or later.  Haven't used it myself, but the documentation claims 
it works.

> If TurboIron is not good idea here then what we should consider as
> alternative ? MLX/RX is not option here as It's too expensive. We need
> just 8 x 10GE ports (best if XFP, but SFP+ is fine with us too).

Big buffers, lots of 10G ports, cheap.  Choose two. :-D

> 4) Other L2 platforms - which other Brocade 1U switch has deep buffers
> and 2x10GE capability ? I also need QinQ, Ingress/Egress policing at
> GE/VLANs ports.

fes-x600 series has 64 megs shared buffers per chassis and has an optional 
2x10GE card.  This works reasonably well for GE access ports with heavy 
outbound traffic (32 megs shared ingress + 32 megs shared egress - i.e. 
250ms line rate buffering best case).  It's a 2U switch though.

Nick



More information about the foundry-nsp mailing list