1
22.7 Legacy Series / PPPoE optimization for 10GbE
« on: December 17, 2022, 10:38:12 pm »
Hi guys,
I know the title can be funny, but here in Italy our ISP TIM has launched XGS-PON 10/2 with legacy PPPoE access.
At the moment I’ve a VM with OPNSense running into a Proxmox 7.2 host (i5 12600k as CPU), this one has 4 vCPU (that I’ve upgraded to 6) and make a speedtest using PPPoE not directly from the itself (so doing NAT) gives me only 6.3/6.5Gbps of throughput.
I’ve switched usual tunables: net.isr.dispatch to “deferred”, bind threads to 1 and max threads to -1, but looks like that cannot surpass that speed. Line obviously will never reach 10Gbps, max speed (with overhead and ISP configuration) is about 8.5Gbps and most of other users that has this type of line can reach a maximum speed of 7.9/8.0Gbps.
In meanwhile i’ve attempted a test (on the same PVE host) using an OpenWRT VM in this two scenarios:
- first test: 8 vCPU, PPPoE and NAT, I can reach without any problem 7.3Gbps (that looks like the maximum that my OLT can reach) from a VM behind.
- second test: 4 vCPU, Double NAT (PPPoE done by ISP router), in this way same speed can be achived from the VM
all tests above were done with stock configuration on the OS (no optimization done)
Any suggestion how to improve situation with PPPoE on OPNSense VM? I want to have public IP directly on the box and avoid double NAT, and in the next few days remove the ISP router and use an XGSPON SFP ONT.
Thanks in advance
I know the title can be funny, but here in Italy our ISP TIM has launched XGS-PON 10/2 with legacy PPPoE access.
At the moment I’ve a VM with OPNSense running into a Proxmox 7.2 host (i5 12600k as CPU), this one has 4 vCPU (that I’ve upgraded to 6) and make a speedtest using PPPoE not directly from the itself (so doing NAT) gives me only 6.3/6.5Gbps of throughput.
I’ve switched usual tunables: net.isr.dispatch to “deferred”, bind threads to 1 and max threads to -1, but looks like that cannot surpass that speed. Line obviously will never reach 10Gbps, max speed (with overhead and ISP configuration) is about 8.5Gbps and most of other users that has this type of line can reach a maximum speed of 7.9/8.0Gbps.
In meanwhile i’ve attempted a test (on the same PVE host) using an OpenWRT VM in this two scenarios:
- first test: 8 vCPU, PPPoE and NAT, I can reach without any problem 7.3Gbps (that looks like the maximum that my OLT can reach) from a VM behind.
- second test: 4 vCPU, Double NAT (PPPoE done by ISP router), in this way same speed can be achived from the VM
all tests above were done with stock configuration on the OS (no optimization done)
Any suggestion how to improve situation with PPPoE on OPNSense VM? I want to have public IP directly on the box and avoid double NAT, and in the next few days remove the ISP router and use an XGSPON SFP ONT.
Thanks in advance