Skip to content

Bug: Ingress Port playing a role in egress member selection in ECMP #29592

Description

@saravanan-nexthop

Is it platform specific

broadcom

Importance or Severity

High

Description of the bug

We investigated the issue of the ingress port playing an unexpected role in ECMP egress member selection on our Q3D T2 testbeds.

What we verified on live hardware is that the ECMP hashing behavior is dependent on the 16-bit hash seed stored per (core, PP-port) in IPPB_LB_SEED_MAP_MEM. When different ingress ports have the same seed, the hashing is exactly the same—an identical 5-tuple egresses the same ECMP member regardless of the ingress port or core. Conversely, when the ingress ports have different seeds, the hashing is different, causing the same flow to split deterministically across different ECMP members.

One possible root cause for this ingress port dependence is a non-uniform programming of this seed table due to a defect in the SAI.
BCM CSP CS00012479434

Steps to Reproduce

Same 5 tuple traffic from different ingress port takes different egress port on Q3D

Actual Behavior and Expected Behavior

Same 5 tuple traffic from different ingress port needs to egress out on same port on Q3D

Relevant log output

Output of show version, show techsupport

Attach files (if any)

No response

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions