Download below 10 Mbits with async download speeds on DSL and Cable Multipath

[3dzug]

I use two different isp with very different connection speeds.
My setup is:

eth0 - local
eth1 - DLS Down 80 Mbit / Up 35 Mbit
eth2 - Cable Down 1000 Mbit / Up 50 Mbit

The aggregation of the upload speed works fine and is never an issue and very satisfyling. Download is not.
Download speed is most of the time 10 mbit or less and heavily fluctuating, sometimes 200 mbit.
If I disable multipath on eth1 the connection speed is stable at 1000 / 50.
As soon as I activate it again on eth1 the speed is 15 down and 85 up.

I am using a mini computer with Intel N 100. Systemload is when running eth2 speedtest not more than 30%. When I do speedtests with multipath on for both and the speed drops to 10 mbit its 2%.
My guess here is thats the download ratio 1:10 here more than a hardware issue.
I tried various things in the setup without success (also VPN for mtcp on eht2 without any impact).

How can I fix that issue?
I'd be happy two ways.

1. As my intention is the aggregated upload it would be good if I can setup eth2 for download only and eth1 and 2 for upload so that the result is 1000 / 85.

2. Although 80 mbit on 1000mbit added is almost not noticable it would be cool for the nerd factor to really get the advantage of 1080 down and 85 up.

What can I do? Setup is attached.

[Ysurac]

Can be an issue on the used VPS.
You can test speed using omr-test-speed on the VPS.
This can also be related to speed between VPS and connection, you can test it with omr-test-speed-server vps eth1 and ``````omr-test-speed-server vps eth2(this is only an information and not a really good speed test).omr-iperf``` can also help as it use iperf server installed on VPS.

[3dzug]

Here are the results:

omr-test-speed
Best server is http://speedtest.frankfurt.linode.com/garbage.php?ckSize=10000, running test: % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 27.0M 0 27.0M 0 0 227k 0 --:--:-- 0:02:01 --:--:-- 1535k

omr-test-speed eth1
root@OpenMPTCProuter:~# omr-test-speed-server vps eth1 Get API token on vps Disable SQM bandwidth limit on eth1 Download test via server vps: % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 1024M 100 1024M 0 0 9076k

omr-test-speed eth2
root@OpenMPTCProuter:~# omr-test-speed-server vps eth2 Get API token on vps Disable SQM bandwidth limit on eth2 Download test via server vps: % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 1024M 100 1024M 0 0 104M 0 0:00:09 0:00:09 --:--:-- 110M

If I do the test the other way enabling eth1 only via vps and disable eth2 I get the maximum speed on eth1 too.

[muko82]

Hetzner VPSes tends to be cutted in bandwidth when they detect that consumer services are used. You have to try many times to find "perfect" settings if you plan to use them

[3dzug]

Do you have a suggestion for another vps service? Speed 1.2 GBit would be nice.
Same issue with HT-hosting. Upload is fine, download is 4-10.

Can I enable somehow upload eth1 and eth2 but download eth1 only?

[KaWiSo]

Do you have a suggestion for another vps service? Speed 1.2 GBit would be nice. Same issue with HT-hosting. Upload is fine, download is 4-10.

Can I enable somehow upload eth1 and eth2 but download eth1 only?

I am quite satisfied with the cheapest webtropia cloud vps, although it is only 1Gbit.
I would always prefer location and therefore rtt, so if you live near Düsseldorf webtropia or servdiscount should be a good option as the servers are hosted by WIIT, formerly myloc.

You could also try using IONOS, their vps are hosted in Berlin I think.

[3dzug]

Thanks, however what I still find odd that I sometimes get the aggregated speed on download but always on upload. Assuming the provider is blocking the service it neither should work any direction.
So what are my options now? I tried three services, (Arovo, Hetzner, HT-hosting) each of it had issues, either slow download or slow upload.

[Ysurac]

Did you try also with SQM disabled ?

[3dzug]

Hi,Upload is less, download same issue .Am 04.11.2024 um 20:00 schrieb Ycarus (Yannick Chabanois) ***@***.***>: Did you try also with SQM disabled ? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[darkman1983]

I had this problem too.
I changed proxy to xray VLESS and download problems gone away. If you use Shadowsocks Rust 2022, it might be related to that.
If i use Shadowsocks RUST 2022, also the configured MPTCP Settings are ignored.
It always uses both WAN's in some sort of load-balancing, also if you set: Master Interface to - on wizard change and MPTCP Scheduler eg. to first, which should give full load to WAN1, until it's saturated and then use second line.

Another observation is: I had that random ping spikes with Shadowsocks RUST 2022, like every 10 to 30 seconds ping over 100 or more. With xray that has also gone away.

With Shadowsocks RUST 2022 it always uses both. If you use xray VLESS, you should see, that 90% of time WAN1 is used until saturated.

Maybe there is an issue with the settings or with the patch applied to shadowsocks (if there is any).
But for xray VLESS, it is recommended, that you have good hardware.

I add another one: For VPN (for UDP traffic), only Glorytun TCP is working the right way.
If i use OpenVPN Bonding, it always shows offline, but it is somehow working, maybe only one route for only one WAN is working here.
Glorytun UDP works like 1-2 hours, then loosing connection and never reconnects.

There seems to be some sort of packetloss ongoing with OMR, it was worse with Shadowsocks RUST 2022, got better with xray, but not full away. I haven't managed to verify, where it comes from. Maybe it's some sort of connection loss to VPS, omrvpn is restarted several times a day. I tested each WAN without aggregation, and there's no packetloss problem on each of them. Only on aggregation.

Maybe my obervations help here...

P.S. I'm using a micro compliance with N100 too.

[3dzug]

@darkman1983 YOU MADE IT <3 ! Its working. I changed proxy to xray VLESS. Connection is stable and I have Down 1000 / UP 75 - 80. Where I got Hetzner running best before it now runs on Arovo, HT-Data and Hetzner similar good. The limit now is my 1GB switch here :).

[a834003831]

I had this problem too. I changed proxy to xray VLESS and download problems gone away. If you use Shadowsocks Rust 2022, it might be related to that. If i use Shadowsocks RUST 2022, also the configured MPTCP Settings are ignored. It always uses both WAN's in some sort of load-balancing, also if you set: Master Interface to - on wizard change and MPTCP Scheduler eg. to first, which should give full load to WAN1, until it's saturated and then use second line.

Another observation is: I had that random ping spikes with Shadowsocks RUST 2022, like every 10 to 30 seconds ping over 100 or more. With xray that has also gone away.

With Shadowsocks RUST 2022 it always uses both. If you use xray VLESS, you should see, that 90% of time WAN1 is used until saturated.

Maybe there is an issue with the settings or with the patch applied to shadowsocks (if there is any). But for xray VLESS, it is recommended, that you have good hardware.

I add another one: For VPN (for UDP traffic), only Glorytun TCP is working the right way. If i use OpenVPN Bonding, it always shows offline, but it is somehow working, maybe only one route for only one WAN is working here. Glorytun UDP works like 1-2 hours, then loosing connection and never reconnects.

There seems to be some sort of packetloss ongoing with OMR, it was worse with Shadowsocks RUST 2022, got better with xray, but not full away. I haven't managed to verify, where it comes from. Maybe it's some sort of connection loss to VPS, omrvpn is restarted several times a day. I tested each WAN without aggregation, and there's no packetloss problem on each of them. Only on aggregation.

Maybe my obervations help here...

P.S. I'm using a micro compliance with N100 too.

"I have already tested in China, and the download speed is still slower than the upload speed. The VPS is hosted on Alibaba Cloud with 100M bandwidth. What settings do I need to adjust?"

[a834003831]

I had this problem too. I changed proxy to xray VLESS and download problems gone away. If you use Shadowsocks Rust 2022, it might be related to that. If i use Shadowsocks RUST 2022, also the configured MPTCP Settings are ignored. It always uses both WAN's in some sort of load-balancing, also if you set: Master Interface to - on wizard change and MPTCP Scheduler eg. to first, which should give full load to WAN1, until it's saturated and then use second line.

Another observation is: I had that random ping spikes with Shadowsocks RUST 2022, like every 10 to 30 seconds ping over 100 or more. With xray that has also gone away.

With Shadowsocks RUST 2022 it always uses both. If you use xray VLESS, you should see, that 90% of time WAN1 is used until saturated.

Maybe there is an issue with the settings or with the patch applied to shadowsocks (if there is any). But for xray VLESS, it is recommended, that you have good hardware.

I add another one: For VPN (for UDP traffic), only Glorytun TCP is working the right way. If i use OpenVPN Bonding, it always shows offline, but it is somehow working, maybe only one route for only one WAN is working here. Glorytun UDP works like 1-2 hours, then loosing connection and never reconnects.

There seems to be some sort of packetloss ongoing with OMR, it was worse with Shadowsocks RUST 2022, got better with xray, but not full away. I haven't managed to verify, where it comes from. Maybe it's some sort of connection loss to VPS, omrvpn is restarted several times a day. I tested each WAN without aggregation, and there's no packetloss problem on each of them. Only on aggregation.

Maybe my obervations help here...

P.S. I'm using a micro compliance with N100 too.

The upload speed can occasionally exceed 100Mbps, but the download speed is generally below 50Mbps. Additionally, during speed tests, the VPS briefly disconnects, showing the message: "Connection lost."


[darkman1983]

I had this problem too. I changed proxy to xray VLESS and download problems gone away. If you use Shadowsocks Rust 2022, it might be related to that. If i use Shadowsocks RUST 2022, also the configured MPTCP Settings are ignored. It always uses both WAN's in some sort of load-balancing, also if you set: Master Interface to - on wizard change and MPTCP Scheduler eg. to first, which should give full load to WAN1, until it's saturated and then use second line.
Another observation is: I had that random ping spikes with Shadowsocks RUST 2022, like every 10 to 30 seconds ping over 100 or more. With xray that has also gone away.
With Shadowsocks RUST 2022 it always uses both. If you use xray VLESS, you should see, that 90% of time WAN1 is used until saturated.
Maybe there is an issue with the settings or with the patch applied to shadowsocks (if there is any). But for xray VLESS, it is recommended, that you have good hardware.
I add another one: For VPN (for UDP traffic), only Glorytun TCP is working the right way. If i use OpenVPN Bonding, it always shows offline, but it is somehow working, maybe only one route for only one WAN is working here. Glorytun UDP works like 1-2 hours, then loosing connection and never reconnects.
There seems to be some sort of packetloss ongoing with OMR, it was worse with Shadowsocks RUST 2022, got better with xray, but not full away. I haven't managed to verify, where it comes from. Maybe it's some sort of connection loss to VPS, omrvpn is restarted several times a day. I tested each WAN without aggregation, and there's no packetloss problem on each of them. Only on aggregation.
Maybe my obervations help here...
P.S. I'm using a micro compliance with N100 too.

The upload speed can occasionally exceed 100Mbps, but the download speed is generally below 50Mbps. Additionally, during speed tests, the VPS briefly disconnects, showing the message: "Connection lost."

Sorry dude, i have just realized, that the other guy is from germany too and i had a similiar problem. I don't know the specifics for China, nor Alibaba. If my observations doesn't help you, maybe you have another problem here.
We changed nothing more than the proxy that is used.

In general, i think the problem started from around october 13 snapshot.