[c-nsp] Cisco Nexus as MetroE switch?

[Tom Hill]

On 03/11/15 11:28, Marian Ďurkovič wrote:
> The reason for these issues was however completely different.
> 
> For the TRILL case, the whole problem was, that ASIC uses L2 headers by default.
> Thus the fix was trivial - one ASIC register was set to non-default value and
> the switch started to load-balance according to L3 & L4 headers in HW natively,
> without any special provisions.

Ahh.. Thank you for the explanation. My vendor didn't tell me what the
difference was, but I knew both had issues. I'm glad your vendor found a
way around it - and it isn't too terrible (certainly nicer than other
hacks). :)

> TRILL switch supports QinQ just fine. If the E-LINE customer wants to use 
> its own VLAN tags, he will just send tagged packets to TRILL switch, which
> will add a second tag and then encapsulate the frame into TRILL container. 
> 
> All you need to do is to configure customer-facing ports on TRILL switch as
> dot1q-tunnel. 
> 
> As mentioned before, we’re using it in Bratislava MAN of academic network
> without any problems (on Trident+)

Oh sure, this is fine if you're happy with QinQ levels of
"encapsulation" for E-LINE services, but then it's down to the vendor to
do L2PT as they see fit. This is where just double-tagging for E-LINE
gets horrid, especially in carrier-on-carrier situations.

For simple tagged access aggregation back to a virtual CPE, or similar,
I'm sure it's pretty good.

> OK, let’s take Cisco as an example. It indeed seems that their FabricPath
> (proprietary TRILL implementation incompatible with RFC6325) is no
> longer promoted and VXLAN+BGP is the new buzzword. But wait, multiple
> posters mentioned here, that it’s even more complex than MPLS.
> Is the ever-increasing complexity really something that customers want?

Customer's don't care until it *breaks*. If we assume complexity ==
unreliability, then maybe this could be true, but I'm fairly sure bad
design (or not fully understanding the vendor's design) is more likely
to contribute to a bad customer experience... This is a bit tangential
though!

> These limitations of Trident ASICs apply to all L2 encapsulations, i.e. TRILL,
> VXLAN, SPB, VPLS… The problem is that Trident can’t preform L2 decapsulation
> and L3 routing in a single pass, so packets need to be recirculated if both
> functions are needed.

Not sure E-LINE really calls for L3 demarcation, but there's definitely
an issue there for some requirements. I'm sure Cisco would love to sell
you A9K for that "problem". :)

-- 
Tom