Slow upload speed to international / far away servers

Hello. I started testing my bpi r4 yesterday but I’m facing slower uploads to international / far away servers. I currently have a 1gb symmetric up and down contract. It’s connected via a Leox SFP SFP attached to a managed switch and the connected to the router via Dac cable to Intel x520.

With this configuration I get 940mbps down and around 850 to 900 up to most to international locations where i have backups and other services. Wit the banana PI r4 I can barely hit half of that. I tried turning on and off flow control directly on Luci and directly on devices configuration ( both SFP+ ports ) and nothing. It did improve a bit using software flow control instead but nothing close of what I had before. It gets even worse if I try to sue WireGuard or tailscale. It’s doesn’t go beyond 200mbps when with x86 I get to 800mbps. My x86 config is on a VM on proxmox using 4 cores on a Intel 12600h and cpu barely moves so not exactly a power horse.

If anyone has any idea how to improve this it would be appreciated.

P.S. this is all wired. I don’t have any WiFi module installed. Just the board and an nvme

Are jumbo frames enabled?

Nope should I ? Where exactly shoud I enable them ? On the wan interface ?

Just google it, or mtu. It is only an idea. Do not know what else you can do, if it goes wrong far beyond your system.

It’s definitely something with the R4 since it doesn’t happen with my x86 vm

But I’ll try the jumbo frames later

But the R4 has no problems with nearby servers, right?

Nope. Just far away ones. It’s like it doesn’t have enough power to process packages. But I don’t see nothing saturated. Same as x64

This definitely does not seem like an R4 issue, but more likely a routing issue with your ISP. I’d try using mtr to see if there are any choke points with your ISP’s peering.

Hello. Thanks for your reply. I really don’t understand how this might be a problem with my ISP when i jjust said that if unplug the two sfp cables connected to the Bpi r4 and conntect it to a X86 vm running on proxmox with nic passthrough the problem doesn’t exist. Nothing else changes.

Is it possible that different DNS servers and hence possibly different counterparties are used?

After all, it’s always DNS.

Nope. I’ve imported the x86 config. Just changed the interfaces which are slightly different.

I suspect I has something to do with hw offloading which is it seems not be properly / totally implemented on the r4. Because if I disable hw offloading and use software and it improves things a bit but cpu can’t handle it apparently !

Do you have the two sfp’s in 1 bridge or how do they connect/forward?

One sfp is connected to a switch where my ISP onu is.

The other sfp is used to Connect to my internal lan.

The lan sfp is on brlan to one of the SFP ports . I’ve tried to install Bridger to see if anything changed and nothing.

The only thing I didn’t try was using bridge at all for. But never thought of it since it wasn’t an issue on x86

As said before before nothing changed from my x86 config ! I just unplugged the two cables from proxmox vm with nic pass through and plugged into the R4.

Well I can tell you is that mine happily routes 1.5, possibly 2gbit/s even transatlantic (openwrt 24.10). I had tested that a few weeks back with a wireguard ProtonVPN endpoint in the US. With all the CDN and anycast stuff going on these days, testing long distance links is actually not entirely trivial anymore.

Pretty sure it’s not the router itself (also no obvious way how it could be). Might be a config issue, MTU would be a prime suspect. Or some non standard routing setup.

That’s not very clear, but I can assume the 2 sfp’s are not both on brlan or brwan, but one is on brlan and another is not, somewhere else?

It all depends what you need the R4 to do, bridging the interfaces, or ip-forwarding (with masquerade)?

First try without offloading, 1gbps should be no problem for the cpu, so something else is going wrong here… Perhaps indeed the routing setup? ‘Far away’ ip numbers are being send in the wrong direction?

This is all very fine but why does it work fine with x86 and mt with r4? I even tried basic config just my wan internet vlan and simple lan

Just brlam with eth2. No brwan

Can you please try access to you wan server from r4 first without any bridge?

Ill try during the weekend. But if thats the issue it’s a complete unpratical bug because i need the 1gb ports on the same subnets as the sfp lan port and i can’t achieve that with out e bridge.

Btw maybe it’s worth mentioning I’m using vanilla openwrt. I tried both 24.10.1 and snapshot.

We have to reduce network complexity to find out where the issue is located, so first test directly on interface without forwarding and virtual intefaces