On Mon, Aug 14, 2023 at 02:07:12AM +0000, Jason Tubnor wrote: >> Hi,
> Not sure how this can happen. Have you destroyed and recreated the interface > in between? Can you easily reproduce this? No I didn't it just seem to drop. It happened twice yesterday but I have, even under continous load, not been able to get it to error again. > I have added a bit more info to the error message, it now also prints the > iface id and the errno. It would be useful if you can reproduce it with those. Thanks and patched across all 3 machines. Below are metrics/results from today's testing: Spoke A: -------- CPU0 states: 1.0% user, 0.0% nice, 27.5% sys, 5.9% spin, 2.9% intr, 62.7% idle CPU1 states: 0.0% user, 0.0% nice, 26.5% sys, 4.1% spin, 0.0% intr, 69.4% idle PID USERNAME PRI NICE SIZE RES STATE WAIT TIME CPU COMMAND 45123 root 2 0 1780K 2864K sleep/0 kqread 8:39 20.75% iperf3 | | \|/ Hub: ---- CPU0 states: 0.0% user, 0.0% nice, 37.4% sys, 16.2% spin, 1.0% intr, 45.5% idle CPU1 states: 0.0% user, 0.0% nice, 40.2% sys, 10.8% spin, 0.0% intr, 49.0% idle | | \|/ Spoke B: -------- CPU0 states: 0.0% user, 0.0% nice, 23.0% sys, 1.0% spin, 6.0% intr, 70.0% idle CPU1 states: 0.0% user, 0.0% nice, 30.4% sys, 2.0% spin, 0.0% intr, 67.6% idle PID USERNAME PRI NICE SIZE RES STATE WAIT TIME CPU COMMAND 32297 _iperf3 2 0 1776K 2688K sleep/0 kqread 4:24 10.55% iperf3 Performance of sending Spoke A through to Spoke B via Hub (iperf3 -t 3600): [ ID] Interval Transfer Bitrate Retr [ 5] 0.00-3600.00 sec 117 GBytes 280 Mbits/sec 85 sender [ 5] 0.00-3600.01 sec 117 GBytes 280 Mbits/sec receiver Performance of sending Spoke A to Hub (iperf3 -t 300): [ ID] Interval Transfer Bitrate Retr [ 5] 0.00-300.00 sec 37.9 GBytes 1.08 Gbits/sec 17 sender [ 5] 0.00-300.00 sec 37.9 GBytes 1.08 Gbits/sec receiver Hub load while under iperf3 test with more throughput above: CPU0 states: 0.0% user, 0.0% nice, 47.0% sys, 6.0% spin, 1.0% intr, 46.0% idle CPU1 states: 0.0% user, 0.0% nice, 46.6% sys, 4.9% spin, 0.0% intr, 48.5% idle Memory: Real: 23M/273M act/tot Free: 176M Cache: 89M Swap: 22M/256M PID USERNAME PRI NICE SIZE RES STATE WAIT TIME CPU COMMAND 1205 _iperf3 2 0 1016K 1736K sleep/1 kqread 0:05 11.43% iperf3 ---- Is there a way to improve the forwarding speed at the hub? Using the same IKEv2 configuration with a VXLAN/VEB/VPORT configuration, I can get ~2.5x the performance from spoke to spoke. Spoke to Hub performance is on par. Cheers
