I have recently been setting up 10Gbps for two Proxmox nodes. I have found that when using Proxmox 6.2 on both machines, the lower powered machine is only capable of sustaining speeds of around 5.8Gbps when on the receiving side (`iperf3 -s`) of an iperf test. When that host is on the sending side, the speed is consistently around 9.4Gbps.
At first I would have possibly attributed the difference to the hardware of the lower powered machine, but then I ran the same tests using a Debian Live 10.5 USB on both hosts. With that setup, the same hardware was able to sustain speeds of around 8Gbps and 9.4Gbps, so around a 2Gbps improvement on the receiving side of the lower powered host.
One thing I noticed when using the Proxmox install, during only one of the iperf runs, the bitrate started at around 9.4 and then quickly fell to its normal 5.8Gbps.
I have tried several options I've found online to get the speeds on the Proxmox install to match that of the plain Debian install, but have not been able to get quite as high of speeds.
Setting txqueuelen to 10000 (`ip link set dev eno3 txqueuelen 10000`) improves things slightly, as does setting both the rx/tx ring params to 4096 (`ethtool -G eno3 rx 4096 tx 4096`).
Iv'e tried changing the tx, rx, tso, gso with ethtool, but any changes I've made there have only decreased the bitrate. Currently I'm using the following ethtool config: `ethtool -K eno3 tx on rx on tso on gso on lro off`
I checked the output of `ethtool -k` on both Proxmox and Debian Live and they were identical. Also worth noting that in both cases, LRO was disabled.
The hardware I'm using, in case it matters:
Node 1: Supermicro X11SSH-F motherboard with Xeon E3-1275 CPU, and an X520-DA2
Node 2: A2SDI-H-TP4F motherboard with integrated Intel Atom C3958 and integrated SoC 10Gbps SFP+
10Gbps SFP+ switch
Is there anything else I should try, or any other settings I should compare in the two setups to match the bitrate from the Debian Live USB on the Proxmox install?
Thanks in advance for any help!
At first I would have possibly attributed the difference to the hardware of the lower powered machine, but then I ran the same tests using a Debian Live 10.5 USB on both hosts. With that setup, the same hardware was able to sustain speeds of around 8Gbps and 9.4Gbps, so around a 2Gbps improvement on the receiving side of the lower powered host.
One thing I noticed when using the Proxmox install, during only one of the iperf runs, the bitrate started at around 9.4 and then quickly fell to its normal 5.8Gbps.
I have tried several options I've found online to get the speeds on the Proxmox install to match that of the plain Debian install, but have not been able to get quite as high of speeds.
Setting txqueuelen to 10000 (`ip link set dev eno3 txqueuelen 10000`) improves things slightly, as does setting both the rx/tx ring params to 4096 (`ethtool -G eno3 rx 4096 tx 4096`).
Iv'e tried changing the tx, rx, tso, gso with ethtool, but any changes I've made there have only decreased the bitrate. Currently I'm using the following ethtool config: `ethtool -K eno3 tx on rx on tso on gso on lro off`
I checked the output of `ethtool -k` on both Proxmox and Debian Live and they were identical. Also worth noting that in both cases, LRO was disabled.
The hardware I'm using, in case it matters:
Node 1: Supermicro X11SSH-F motherboard with Xeon E3-1275 CPU, and an X520-DA2
Node 2: A2SDI-H-TP4F motherboard with integrated Intel Atom C3958 and integrated SoC 10Gbps SFP+
10Gbps SFP+ switch
Is there anything else I should try, or any other settings I should compare in the two setups to match the bitrate from the Debian Live USB on the Proxmox install?
Thanks in advance for any help!