[j-nsp] FPC/PFE route scaling

[Thomas Nikolajsen]

>I'm having a hard time understanding route scaling on the bigger M and
MX boxes.

You are not alone.
Official documentation is very sparse in this area.

>In a CFEB based box maximum >FIB is around 500k due to the limited SRAM
on the CFEB holding the FIB.
>On a CFEB-E this is increased.
>
>Now looking at the M320 or MX240 the FIB is present on all FPC boards
in each PFE, correct?

Yes

>What is the architectural difference when using a
>distributed setup, and how does it affect maximum FIB size? Eg. On an
FPC3-E2 in an M320 what can be expected to be the limit?

On M320 maximum FIB size is 1M ipv4 or 768k ipv6 routes.
Expected limit depends on many factors; around 700k ipv4 routes seems
feasible here.
The many factors include use of firewall filters, number of aggregated
interfaces & ECMP.

Also use of features like LFA will influence FIB use somewhat.
http://www.juniper.net/us/en/local/pdf/whitepapers/2000345-en.pdf has
some info on LFA,
it also describes different types of FIB entries;
M320 FPC can accommodate different amount of them (has different jtree
memory use).

On E3 FPCs expected limit can be somewhat higher:
http://www.juniper.net/techpubs/en_US/junos10.4/topics/task/configuratio
n/junos-software-jtree-memory-repartitioning.html

On MX960 (MX240 should be same) maximum FIB size on Trio based PFE is 2M
ipv4 or ipv6 routes.
Expected limit should be much closer to maximum here.

 -thomas