[j-nsp] Experience with MX10003

[adamv0025 at netconsultings.com]

Hey,

> Giuliano C. Medalha
> Sent: Thursday, January 25, 2018 6:43 PM
> 
> We are testing the MX10003 right now.
> 
> It is a pretty amazing box.
> 
> The routing engine is so power for control plane ... bgp, ospf etc
> 
So it doesn't take several minutes to import local routes to a newly configured VRF in a setup with 2000VRFs and 2M of prefixes? 
 
> The timers for BGP to put routes on FIB is something special ( after turbo fib
Actually ib programming optimizations will be introduced to MX as well (some might already have been introduced in latest codes). 

> ).  With BGP PIC reconvergence is very fast too.
> 
Well the sub 50ms with BGP PIC is not optional :) . 

> We will test the mpc for traffic with spirent testcenter.
>  
I'd love to hear whether the pre-classifier fix now available on MX (mpc3-ng and mpc7 anyways) made it into the release for this platform. 
To test this you can overload the PFE and see if it starts dropping EF traffic in WAN input. 
While on that Gen3 in mpc7s had some fabric backpressure bugs, so you might want check if backpressure works correctly on the MX10003 iteration of the Gen3 (the unlocked). 

adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::