Pre-purchase check: real-world experience running 8 × 2-port LACP LAGs to Proxmox/Ceph hosts on a CRS326-24S+2Q+RM (bond hardware offload, 1-sec LACP, VLANs + MTU 9000, buffer depth under Ceph bursts, fan noise).

Hardware MikroTik hardware general
1

I’m looking at a CRS326-24S+2Q+RM to replace a noisy Cisco 4500X in a home lab.

Setup

  • 3× Proxmox VE nodes running Ceph, all bonds 802.3ad (LACP) with lacp_rate fast on Linux.
  • 8 × 2-port 802.3ad bonds in total (16 SFP+ ports, all DAC): 3 for Ceph, 3 for management/migration, 1 for VM traffic, and 1 uplink to a Cisco 3850.
  • Single switch, no MLAG. Pure L2: access VLANs plus one trunk. Routing stays on my router; the switch only has a management IP.

Questions

  1. 8 offloaded 802.3ad bonds: I know 802.3ad is hardware-offloaded on CRS3xx. Has anyone run 8 or more 802.3ad bonds on a 98DX8332 with every one showing H in /interface bridge port print, none falling back to the CPU?
  2. LACP fast rate: is lacp-rate=1sec stable against Linux bonding, including after reboots on either side?
  3. VLAN filtering + MTU 9000 on bonds: any gotchas with vlan-filtering=yes, L2MTU and hardware offload together?
  4. Buffers under Ceph bursts: the 98DX8332 seems to have only about 2.6–3 MB of shared buffer. Has anyone seen tx-drops during Ceph recovery/backfill? Did hw-QoS or flow control help?
  5. Fan noise: is it tolerable in a home office?

Thanks!