From the specs, i see that the 3850 only supports 8K of IPv4 routes. Is this per VRF or across all VRFs ?
Isn't this number of 8K routes very limited, especially for the 3850XS models which are targeted at core/aggregation level ?
The switch supports up to 27 VRFs, but this means only 300 IPv4 routes per VRF and if you are mixing in with IPv6, even less.
We have VRFs of around 2500 route entries. It would mean we can only support 3 VRFs ? (and this without any IPv6 deployment)
where are you seeing that ? im looking at the data sheet and its up to 24000 routes ipv4
it depends on what the sdm is set to as well and what license is in place
looking at 1 of my 38s running ip services with sdm set to advanced it can take up to 24
sh sdm prefer
Showing SDM Template Info
This is the Advanced (high scale) template.
Number of VLANs: 4094
Unicast MAC addresses: 32768
Overflow Unicast MAC addresses: 512
IGMP and Multicast groups: 8192
Overflow IGMP and Multicast groups: 512
Directly connected routes: 16384
Indirect routes: 7168
Security Access Control Entries: 3072
QoS Access Control Entries: 3072
Policy Based Routing ACEs: 1024
Netflow ACEs: 768
Wireless Input Microflow policer ACEs: 256
Wireless Output Microflow policer ACEs: 256
Flow SPAN ACEs: 512
Control Plane Entries: 512
Input Netflow flows: 8192
Output Netflow flows: 16384
SGT/DGT entries: 4096
SGT/DGT Overflow entries: 512
These numbers are typical for L2 and IPv4 features.
Total number of IPv4 routes (ARP plus learned routes)
"Directly connected routes" are local VLANs and loopbacks. That are not "learned" routes, ie OSPF or BGP learned routes. These are "Indirect routes", and this limit is 7168. Or am i wrong ?
The original 3750G-12S, designed for distribution, had a larger TCAM and additional SDM templates. I can only guess that Cisco found having a "special" variant wasn't worth the trouble or they decided they didn't want to compete against other products in their line (existing or planned) that would be more suitable in such a role for larger networks. Either might explain why the 3860XS is as it is.
I haven't any direct experience with the3650/3850 series, but assuming its architecture is somewhat similar to earlier 3560/3750 series, I would expect the number of routes supported to be a global resource, i.e. I would expect it to be shared across VRFs.
Although Mark's references notes IPv4 routes as 24K, table 1 in http://www.cisco.com/c/en/us/products/collateral/switches/catalyst-3850-series-switches/deployment_guide_c07-727067.html, lists
Maximum supported indirect routes
This document also notes:
Table 1 details the resource allocation for VLAN and advanced SDM templates. These resource allocations are based on L2 and IPv4 features. Because IPv6 features consume twice the ternary content addressable memory (TCAM) table size of IPv4 table entries, the switch supports half the number of TCAM table entries for IPv6.
In my company, we didn't move from 3750s to 3850s for similar concerns especially as we were looking to roll out IPv6.
I am running Denali 16.3.3 with 2 VRF of each around 2100 IPv4 external prefixes:
It doesn't fit 100%, only 3500 used instead of 4200, but 3500 is still already 50% of available space.