Latency of Speedfusion Asymmetrical Bonding?

Does asymmetrical routing with speedfusion (upload only via WAN1 and download only on WAN2) add latency?

I did the following ping tests at the same time to my fusionhub’s public IP address:

Assymetrical Routing speedfusion (Starlink download Verizon Upload)
Minimum = 73ms, Maximum = 144ms, Average = 99ms

Starlink Only (no speedfusion)
Minimum = 58ms, Maximum = 120ms, Average = 70ms

Verizon Only (no speedfusion)
Minimum = 68ms, Maximum = 145ms, Average = 88ms

I did another test with speedfusion running pinging the same IP address in the same city as the fusionub with 100 pings:

Assymetrical Routing speedfusion (Starlink download Verizon Upload)
Minimum = 80ms, Maximum = 133ms, Average = 103ms

Starlink Only (speedfusion, verizon wan disconnected)
Minimum = 65ms, Maximum = 93ms, Average = 74ms

Verizon Only (speedfusion, starlink wan disconnected)
Minimum = 74ms, Maximum = 123ms, Average = 92ms

Speedfusion has FEC=off, no ms buffer specified, no wan smoothing, wan1 upload only, wan2 download only.

Performance is excellent in terms of download and upload speed with speedfusion set to asymmetrical bonding, but I was hoping latency would not be greater than the highest latency individual connection (and hoping it might be a little better than the higher latency individual link) since it is doing half the route on one connection and the other half of the route on the other (I thought.)

So I was surprised that average speedfusion assymetrical latency is about 12% higher than the lower latency WAN individually and 40% higher than the higher latency individual WAN.

Is this expected latency, or should I do some more troubleshooting/tweaking to try and improve this?

It feels just a little bit more laggy with speedfusion active in this manner than without it, but I like the download/upload file speed performance with it set to asymmetrical.

Latency in SpeedFusion is normalised effectively to that of the active link with the highest latency, so that behaviour is expected.

You are dealing with two WANs here that are both going to be somewhat variable in terms of base latency, latency under load and also jitter so the results are always going to be somewhat variable.

If the primary goal here though is outright throughput across the PepVPN configuring asymmetric links is a good way to go about it, I’d also possibly experiment a bit with the DWB algorithm as another option as under sustained use I have found it to work very well in these kinds of setups.

1 Like

Are there ways to push the DWB algorithm towards certain connections? My Starlink tends to get weighted away a bit too aggressively…