Recently i did change my setup for Internet access.
Previously i was connected behind a bridged provider Box through an RB3011 Ether1 interface, without VLAN. Using RouterOS 6.49.11.
The Internet bandwidth was 940 mb/s download, 550 mb/s upload, for IPv4 and IPv6. It was the normal bandwidth of this provider.
Then i decided to remove the provider box that was noisy, not useful and power consuming. Then i did connect the fiber ONT module (GPON to Ethernet converter) to the SFP+ interface on the RB3011 through the same DAC cable that was connected before to the provider box.
I needed to change only a few settings on the RB3011 :
- changed MAC address of the SFP+ interface to spoof the provider box Wan mac address
- disabled the SFP+ interface Auto Negotiation and set it to 1 Gbps mode.
- add a VLAN interface on top of the SFP+ interface, because the provider Internet traffic is on VLAN 836
- rise L2MTU of the SFP+ interface to 1704, MTU to 1700, rise MTU of the VLAN 836 interface to 1700 (to make room for an IPIPv6 tunnel)
- modify IPv6 public address and default IPv6 route to get access to Internet without using the provider box as a gateway
- add an IPIPv6 tunnel on top of the VLAN 836 interface to get access to IPv4 Internet (this provider, named FREE, is IPv6 native but need a tunnel for IPv4 delivery)
Everything else did stay identical.
After doing that, i got a normal IPv4 and IPv6 connectivity, without packet drop, but the bandwidth did slow down to about 1/3 the original bandwidth.
I have now around 350 mb/s download and 370 mb/s upload with 75% to 80% CPU consumption. It is not a IPIPv6 tunnel problem because IPv4 and IPv6 bandwidths are almost the same.
It seems that the only thing that could have slow down things is the VLAN 836 interface on top of the SFP+ interface.
Does someone have some background on this problem, or an idea to solve it ?
Edit : i've just discovered that i have a large number of RX too long packets during downloads :
Edit : after investigating, this is not reporting receive errors, just packets that are larger than something like 1500 bytes. This is a bug : the counter should take into account the MTU setting of the interface. Obviously it does not take it into account.
Same problem reported here :
viewtopic.php?t=162346
Previously i was connected behind a bridged provider Box through an RB3011 Ether1 interface, without VLAN. Using RouterOS 6.49.11.
The Internet bandwidth was 940 mb/s download, 550 mb/s upload, for IPv4 and IPv6. It was the normal bandwidth of this provider.
Then i decided to remove the provider box that was noisy, not useful and power consuming. Then i did connect the fiber ONT module (GPON to Ethernet converter) to the SFP+ interface on the RB3011 through the same DAC cable that was connected before to the provider box.
I needed to change only a few settings on the RB3011 :
- changed MAC address of the SFP+ interface to spoof the provider box Wan mac address
- disabled the SFP+ interface Auto Negotiation and set it to 1 Gbps mode.
- add a VLAN interface on top of the SFP+ interface, because the provider Internet traffic is on VLAN 836
- rise L2MTU of the SFP+ interface to 1704, MTU to 1700, rise MTU of the VLAN 836 interface to 1700 (to make room for an IPIPv6 tunnel)
- modify IPv6 public address and default IPv6 route to get access to Internet without using the provider box as a gateway
- add an IPIPv6 tunnel on top of the VLAN 836 interface to get access to IPv4 Internet (this provider, named FREE, is IPv6 native but need a tunnel for IPv4 delivery)
Everything else did stay identical.
After doing that, i got a normal IPv4 and IPv6 connectivity, without packet drop, but the bandwidth did slow down to about 1/3 the original bandwidth.
I have now around 350 mb/s download and 370 mb/s upload with 75% to 80% CPU consumption. It is not a IPIPv6 tunnel problem because IPv4 and IPv6 bandwidths are almost the same.
It seems that the only thing that could have slow down things is the VLAN 836 interface on top of the SFP+ interface.
Does someone have some background on this problem, or an idea to solve it ?
Edit : i've just discovered that i have a large number of RX too long packets during downloads :
Edit : after investigating, this is not reporting receive errors, just packets that are larger than something like 1500 bytes. This is a bug : the counter should take into account the MTU setting of the interface. Obviously it does not take it into account.
Same problem reported here :
viewtopic.php?t=162346
Statistics: Posted by FIPTech — Sat Jan 06, 2024 5:57 pm