Clients are unsuably slow (<1mbps/s)

[autobottodoggo]

Server:

• i5-4590
• 16GB RAM
• 2x 500GB Samsung Evo SSDs on RAID controller
• 50m symmetrical fiber WAN
• SoftEther dev (latest version as of a few days ago)
• 2x Gb NICs connected to Gb switch (need the ListenIP param in the dev version to use only one NIC)
• Rocky Linux 8

The clients are all Windows workstations and the specs vary, but they are all on 50m symmetrical WANs. The server is recently rebuilt, was previously running Windows 10 Pro (needed Linux to expand client connection capacity).

The strange thing is, I tested this configuration beforehand (Rocky Linux + SoftEther dev version) on a much lower spec'd system (old Core 2 and one of the NIC's was only 10/100) and the VPN performance was excellent. It would have surpassed 100mbps if not for the 10/100 card.

What is going on here? This makes no sense.

[shakibamoshiri]

First you can test the speed and throughput via GUI or CLI of an SE server and client. (between clients and a server)

Second this might help
https://www.vpnusers.com/viewtopic.php? ... 972#p98675

[autobottodoggo]

I tested with SoftEther's built in performance tool. It reports ~95mbps on average, but that's not what I'm seeing in reality. Using LAN Speedtest lite (just writes and reads a file to a target location), I get ~50mbps write (okay, I guess), but an abysmal 1-2mbps read.

To reiterate, when the server was running Windows, I'd get about 50mbps both ways. On my test Core 2 system with an identical setup (Rocky Linux, SE dev, 2 NICs), it hit nearly 100mbps both ways. Both using LAN Speed Test lite.

I'm not using SecureNAT.

I'll read through that linked post and see what I can find, thanks.

[solo]

I'm not using SecureNAT.

So...

If you are using local bridging to make a bridged connection between a Virtual Hub and a physical LAN as described in section 3.6 Local Bridges, we recommend you set aside a network adapter specifically for this purpose. This will result in the best performance when using local bridging.

Furthermore...

No Protocol Stack is Used for the Local Bridge Network Adapter

Where there is a network adapter prepared on the computer for use exclusively in local bridging, it is recommended that the TCP/IP protocol and other protocol stacks be disabled on that network adapter to enhance performance. The role of the local bridge network adapter is to release Ethernet frames between the Virtual Hub and the physical LAN, entirely without the need for intervention from the protocol stack of the OS running the Virtual Hub.
...
Setting a fixed IP address for the Local bridge network adapter.

For Linux and Solaris, it is possible to use the [ifconfig] command to obtain a result equivalent to assigning an IP address of 0.0.0.0 to the local bridge network adapter.

[autobottodoggo]

If you are using local bridging to make a bridged connection between a Virtual Hub and a physical LAN as described in section 3.6 Local Bridges, we recommend you set aside a network adapter specifically for this purpose. This will result in the best performance when using local bridging.

That's exactly what I did. I have two physical NICs in this system. One is dedicated to VPN. I'm running SE dev version so I can use the ListenIP parameter.

Code: Select all

netstat -lnt

shows the VPN service is, in fact, only listening on the IP assigned to that NIC.

In any case, I just spun up a fresh test system with Rocky Linux, two NICs, SE dev. 76mbps up and down. This is an Core 2 system with really basic hardware, how is it performing so much better? This makes no sense.

Where there is a network adapter prepared on the computer for use exclusively in local bridging, it is recommended that the TCP/IP protocol and other protocol stacks be disabled on that network adapter to enhance performance.

Forgive my ignorance, but how would this even work? If it has no IP, how am I supposed to configure my router's port forwarding?

[solo]

This is an Core 2 system with really basic hardware, how is it performing so much better? This makes no sense.

Different NICs - different drivers. Swap the NICs around for another test.

[autobottodoggo]

Different NICs - different drivers. Swap the NICs around for another test.

Swapped their roles (on my test system). 44mbps up, 40mbps down. Not quite as good, but still acceptable.

Edit: just tested the production server again, now I'm seeing about 44mbps up/down, which is good... except that I changed nothing. The system isn't under heavy load, both NICs are Gb.