Switching & Routing

 View Only

Fabric Extend via L3 MTU Issue

  • 1.  Fabric Extend via L3 MTU Issue

    Posted 11-08-2021 11:03
    Edited by Robert Bastell 11-08-2021 11:12
    I will apologize in advance if there is a better community for this question, I guessed as to where it could go.


    • We have a number of locations which we have extended our Fabric.
      • 21 locations that connect to our CORE via a subscribed L3 service. 
        • The service handles routes to each node by static route definition.
        • The service has a subscribed MTU of 3900.
      • 4 locations that connect to our CORE via dark fiber or a shared L2 service (METRO) where the VLAN ids are unique to the fiber use partners.
    • We've extended the fabric from our CORE (VSP7400s) to each of the locations which can be a VSP4900 or XA14800 depending on that location.
      • All system MTUs are 1950.
    • ISIS Adjacencies are up.
    • ISIS LSDB shows all locations.
    • ISIS SPBM I-SID shows all currently active C-VLANs with I-SIDs.
    • L2 VSN traffic works with the conditions described below.

    Issue Description

    • L2 VSNs between dark fiber/L2 service connected locations have no issue with any flows.
    • L2 VSNs between the L3 subscribed service connected locations have no issue as long as the MTU doesn't exceed a given value
      • Example:
        • (Windows 10 System)
          • [WORKS] ping A.B.C.D -l 1470 
          • [FAILS] ping A.B.C.D -l 1741

    • Verified MTU settings between L3 subscribed service connected locations:
      • (Windows 10 System)
        • [WORKS] ping A.B.C.D -l 1472 -f
          • A.B.C.D is a non-fabric endpoint reached via L3 subscribed service.
    • Removed 2 devices reachable via the L3 subscribed service from our main fabric and created a new one between those devices themselves.
      • Used multiple incrementing values per test for command on both sides of fabric:
        • logical-intf isis 1 dest-ip A.B.C.D mtu ABCD
          • ABCD --> 1950, 2000, 2050
      • Same results.
    • Changed the fabric endpoints to both be XA1480s and recreated a fabric between them.
      • Same results.

    Can someone recommend some testing or have ideas on how to solve this issue? 

    Thank you in advance.