Measuring the impact to user traffic for link failures on the ACX5048 when the router has learned a fairly large number of routes (~22000 FIB routes - note that the node is configured with the lpm-profile so this should not be exceeding the limit) and finding that it is larger than we had anticipated. This has been noticed since moving to Junos 21.2 With little in the way of route load, we're seeing times around the 80-300 ms mark. This is already larger than expected as we were seeing times around 40-80 ms when running Junos 19.4.

All LSPs have fast-reroute enabled so we were really expecting around the 100 ms mark which is generally reached but not always. When we increase the number of routes were seeing some further increase - up to around 500 ms, but disturbingly were were seeing L2VPN and VPLS services experiencing an outage of up to 5 seconds! Looking at the log messages, it seems that the services are actually sometimes going down and then taking seconds to come back up again.

The ACX platform had following limitations:

IPv4 FiB size: 20k

Ipv4 RIB size: 300k

Max number of labels: 3000

Max number of BGP-LU tunnels:  1021

Max number of BGP adjacency:   256

Max number of label stack depth: 3 labels

Max number of MPLS LSP:       2000

Please note, these limitations are expected to be working in ideal condition. In the real network, if the fast-reroute and adaptive (MBB) are enabled, more resource will be used. 

The possible workaround is remove the fast-reroute settings, and check whether adaptive (MBB) can be also removed.

For L2VPN, the following knobs may be useful: 

set protocols bgp group IGP family l2vpn signaling delay-route-advertisements minimum-delay routing-uptime 300

set protocols bgp group IGP family l2vpn signaling delay-route-advertisements minimum-delay inbound-convergence 10