[c-nsp] PE "Sprawl" - P/Core Router suggestions.

[Phil Bedard]

It’s all VoQ based so it’s similar to other VoQ-based hardware, so there is no such thing as an “egress” drop.  There is an ingress traffic manager stage early in the pipeline that does most of the work.  On the fabric based systems there is a request/grant paradigm for fabric access.  At an early point in the pipeline it can do packet replication, which is where mirroring and netflow come into play.  

Phil 

-----Original Message-----
From: <adamv0025 at netconsultings.com>
Date: Friday, December 30, 2016 at 10:55
To: Phil Bedard <philxor at gmail.com>, 'Saku Ytti' <saku at ytti.fi>, 'Tim Durack' <tdurack at gmail.com>
Cc: <cisco-nsp at puck.nether.net>
Subject: RE: [c-nsp] PE "Sprawl" - P/Core Router suggestions.

    > Phil Bedard
    > Sent: Thursday, December 29, 2016 7:30 PM
    > 
    > Afaik, all the Cisco Jericho/Qumram based cards/fixed boxes use the 16MB
    > on-chip and 4GB of external DRAM.  The 4GB is broken up into 1000 byte
    > chunks from what I remember.
    > 
    Hmm 1000 byte chunks? That's not very efficient is it. 
    
    Yes the Cisco incarnation of the DNX/Dune family is using 16M on-chip
    in-flight buffer and 4G off-chip packet delay buffers. 
    But I can't dig out how does the pre-classifier, arbiter and backpressure
    work on these chips.
    I was pleasantly surprised by the pps rate though, round ~80% line-rate for
    64B at L2 is not that bad (it's like ASR9k's Tomahawk NPU). 
    But I would love to know what performance tax will QOS and sFlow incur on
    the Dune chips.   
    
    
    adam
    
    
    netconsultings.com
    ::carrier-class solutions for the telecommunications industry::