Support #369
closedOAM: Switch config fixes (core-sw01 LAG, LACP, member list)
30%
Description
Switch configuration issues identified in proxmox/docs/AUDIT-2026-07-30.md.
Scope (needs approval - switch changes):
- Fix core-sw01 ch1 member list (g16/g17 mismatch, only 3 of 4 links active)
- Fix LAG hash (currently layer-2-3, caps NFS cross-rack at 1 Gbps/pair)
- Convert cross-rack trunk from mode=on to LACP (no failure detection currently)
- Remove stale tsys8 config from tor3-stor
- Investigate port errors on core-sw01 (g9/g38/g26 inErr)
Code: PFVCluster/netinfra/switches/
Updated by Charles N about 15 hours ago
Updated by Charles N about 15 hours ago
- Status changed from New to Resolved
Updated by Charles N about 15 hours ago
- % Done changed from 100 to 50
Update 2026-08-07: pulled core-sw01 running-config via conman. Confirmed 'port-channel load-balance layer-2-3' (the datanet bottleneck). iperf3 showed 4-stream aggregate stuck at ~908Mbps (single link) from tsys7->tsys5 despite tsys5 having 2-port bond. Fix command prepped in netinfra/switches/core-sw01-lacp-hash-fix.cmds (layer-2-3-4, non-disruptive). BLOCKED on onsite console application - conman is read-only. User onsite today.
Updated by Charles N about 15 hours ago
- % Done changed from 50 to 30
Update 2026-08-07: Verified via conman console that Dell PowerConnect 5448 does NOT support layer-2-3-4 load balancing. Available options are: layer-2, layer-2-3, layer-3 only. Current 'layer-2-3' is already the best option. The LACP hash improvement is NOT achievable on this hardware — this is a switch platform limitation, not a config error. Single-client NFS will cap at ~1Gbps per flow regardless; aggregate multi-client traffic DOES distribute across both LACP links with layer-2-3. Remaining #369 items (ch1 member fix, mode=on→LACP) still valid but lower priority.