On 27.7.2021. 17:36, Christopher Sean Hilton wrote: > On Sat, Jul 24, 2021 at 10:24:28AM -0000, Stuart Henderson wrote: >> On 2021-07-23, Christopher Sean Hilton <[email protected]> wrote: >>> On Fri, Jul 23, 2021 at 11:19:35AM -0400, Chris Hilton wrote: > > [ ...snip... ] > >>> >>> Answering my own question, it looks like the Xeon D is intels newest >>> low power stuff. I'll look there. >> >> Not particularly new, Xeon D 1500 series are from 2016 or so and still >> seem to be the range to go for if you care about good power use. Look >> at supermicro X10SDV (Xeon D 1500 series) or M11SDV (AMD EPYC). Sadly >> the M11SDV only has copper nics, X10SDV have decent ix(4) SFP+ plus >> some copper. (X10 is an older supermicro range, I'm not sure what the >> availability is like). >> >> supermicro, if you're reading, an EPYC board with a couple of SFP28 >> onboard would be nice... >> >> Sample dmesg from one of the X10SDV models - em and ix are onboard, >> ixl is a card: >> >> OpenBSD 6.8-current (GENERIC.MP) #220: Thu Dec 10 20:03:29 MST 2020 >> [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP > > [ ...snip ] > > Thanks to everyone for the answers that they provided. Just a late > followup here. I thought through my testing rig and realized that it > was slightly flawed. I was originally using one of the Atoms as an > iperf endpoint. That obviously messed up the tests. I retested using a > pair of machine which I know can saturate a 1Gb/s connection. My > new test rig is a pair of MacBook Pro's with Thunderbolt Ethernet > adapters: > > * With just a GigE switch connecting the test machines, I measured a > transfer rate of 942 Mb/s. The test program was iperf3. > > * With OpenBSD 6.8 running a bridged configuration on an Intel Atom > D525 with internal and external "em" nics, and filtering using pf. > I measured a rate of 775 ~ 850 Mb/s. Again, the test program was > iperf3. >
maybe you can update to snapshot or 6.9 and try veb(4) instead of bridge(4) ? > Testing the routed configuration on my Atom C2758 is a little more > difficult. I'll set that up next week. I expect that the transfer rate > through that combination will be a little lower since routing is more > difficult than bridging. > > I am currently shopping Intel Xeon-D hardware. I plan to eventually > replace the D525 bridge with the C2758 running in a bridged > configuration and use new Xeon-D hardware for the router. > > -- Chris > >
