Sorry, forgot to mention 2.0 nanobsd August 2 snapshot. On Wed, Aug 25, 2010 at 12:20 AM, David Burgess <apt....@gmail.com> wrote: > I'm using a pair of onboard (vr) NICs on a net5501-80 (500 MHz Geode) > with vlans to firewall a 36/4 mlppp connection. During heavy download > top reports interrupts around 40-50% CPU usage with most of the > remainder being idle. > > I dropped in an Intel Pro 1000 GT (em, PCI) in place of one of the > onboards to handle the internal vlans and during heavy downloading the > interrupts dropped down to around 20%, but now the kernel process was > reporting ~17% CPU usage. The idle process was not significantly > different from the vr NIC to the em. > > I was surprised by this result, not only because of Intel's sterling > reputation among pfsense users, but also because of the fact alone > that the Intel NIC is gigabit hardware (on a gigabit switch). > > Was I wrong to expect a drop in CPU usage with the Intel GBE? > > Also, before somebody mentions it, TSO and LRO were enabled for this > test. I tried disabling LRO, but this immediately caused pfsense to > become unresponsive on the network and the serial console. After > resetting it LRO was still enabled, so I didn't provoke it further. > Within a couple hours pfsense had locked up again, so I moved the LAN > cable back to the onboard NIC and it's been running stably for 17 > hours since (with the Intel card still installed but not assigned). > > Thoughts? > > db >
--------------------------------------------------------------------- To unsubscribe, e-mail: support-unsubscr...@pfsense.com For additional commands, e-mail: support-h...@pfsense.com Commercial support available - https://portal.pfsense.org