FreeBSD Friends,
Back on 10.0-RELEASE I learned that using in-kernel NAT with IPFW performed much better after disabling all of my NIC's hardware acceleration features. At the time this meant
Today I'm configuring two bge(4) interfaces with 12.1-RELEASE-p5. Unlike my 10.0-RELEASE days, these interfaces use LACP and tagged VLAN traffic.
The default accelerations look like this:
By supplying
This performance is still OK, but not knowing whether I really needed to disable everything bugs me.
Rob Roy
Back on 10.0-RELEASE I learned that using in-kernel NAT with IPFW performed much better after disabling all of my NIC's hardware acceleration features. At the time this meant
-tso -rxcsum -txcsum -vlanmtu -vlanhwtag -vlanhwtso.Today I'm configuring two bge(4) interfaces with 12.1-RELEASE-p5. Unlike my 10.0-RELEASE days, these interfaces use LACP and tagged VLAN traffic.
The default accelerations look like this:
options=c019b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4,VLAN_HWTSO,LINKSTATE>By supplying
-tso -rxcsum -txcsum -vlanmtu -vlanhwtag -vlanhwtso, I was able to cut the accelerations down to this: options=80080<VLAN_HWCSUM,LINKSTATE> iperf -s and iperf -c tests between this computer and another showed only a 2% throughput loss after switching the stuff off.This performance is still OK, but not knowing whether I really needed to disable everything bugs me.
- Does anybody know whether it's still necessary to disable hardware acceleration features, in general, with IPFW + in-kernel NAT?
- Does anybody know which of these acceleration features I could leave enabled, without interfering with IPFW + in-kernel NAT?
Rob Roy