What we're trying to do:
- track the per-VLAN traffic ingress and egress volumes.
How:
We utilize SNMP from another host that once a minute queries the traffic counters for all interfaces and logs this to rrd files.
The problem:
Whist we are in agreement (from the Switch side) to the traffic on the LACP port back to the switching infrastructure, the traffic volumes for the individual VLAN interfaces does not. We're seeing kbps values (even when live monitoring, perhaps mbps) when we're expecting several hundred mbps and even gbps values.
Our configuration:
Our setup uses HWL3 offloading for IPv4, which is the bulk of our traffic (without this we don't stand a chance of dealing with the relevant traffic volumes).
Code: Select all
/interface ethernet switch
set 0 l3-hw-offloading=yesCode: Select all
/interface bonding
add mode=802.3ad name=bond-cisco slaves=\
sfp-sfpplus1,sfp-sfpplus2,sfp-sfpplus3,sfp-sfpplus4
/interface bridge
add ingress-filtering=no name=bridge-cisco port-cost-mode=short \
protocol-mode=none vlan-filtering=yes
add name=bridge-lo port-cost-mode=short
/interface bridge port
add bridge=bridge-cisco interface=bond-cisco internal-path-cost=10 path-cost=\
10
/interface ethernet switch l3hw-settings
set fasttrack-hw=no
/interface bridge vlan
add bridge=bridge-cisco tagged=bond-cisco,bridge-cisco vlan-ids=\
2-102,104-4094
We would then add VLANs similar to this:
Code: Select all
/interface vlan
add interface=bridge-cisco name=bond-cisco.2-internal-routing vlan-id=2
Once routes are offloaded however what we're seeing is that whilst bond-cisco will show ~2.8Gbps, the sum of the VLANs will only show ~10Mbps.
We take this as the massive BULK of our traffic is getting offloaded (a good thing), but it also means it's extremely difficult to monitor which peering links are congested - especially considering that the switches don't provide /port/vlan SNMP counters either, and in some cases we've got a handful of related VLANs aggregated on the same LACP bundles (typically less than 5, but enough to skew the results of our traffic graphing).
Is there a way to fix this that I don't seem to find? Bug or intended behaviour?
