Need input for configuration of balance 1350 for maximum throughput

[ACTS]

Hey there,

I recently acquired a used Balance 1350 HW2 and am in the process of experimenting with settings. I have been using an openMPTCPRouter, installed on an old workstation of mine, that I have upgraded with a PCIx 4-Port ethernet card to expand the port count.
Both with the openMPTCPRouter and the Balance 1350 I connect to a virtual server that acts as bonding server. (FusionHub Solo and the OMR VPS, 2.5G uplink, 4vCPU, 8GB RAM each)

My use case for that is to bond multiple WAN connections to one faster link, for mobile use in live event production environments or for filmsets to upload data over accumulated upload speeds.
With the openMPTCPRouter this actually worked very well and I was able to add up my single WAN connection speeds without a lot of loss due to tunnel overhead.
In my testing environment I have a landline internet connection with 100/40 and two 5G routers with around -85dBm signal strength, which give me around the same bandwith. sometimes more, if the cell is not heavily loaded. With the openMPTCPRouter I achieved linkspeeds of 350/150 peak performance, which are roughly an addition of my available WAN connection speeds, which is a great result I think.

Now… I struggle to achieve the same thing with my Balance 1350 for now. I experimented a lot and found out already, that the vpn profile on router and server do not have to be configured the same way, which seems obvious to me now, after I found that out
The best results so far I achieved with these settings:

• TCP Ramp up: active
• WAN Smoothing: Off
• Forward Error Correction: Adaptive
• Policy: Dynamic Weighted Bonding
• Congestion Latency Level: Default

When I had the same settings on the server side, link speeds were bad, probably because both sides enforce those rules even though that does not seem to be neccessary, because it already is a bidirectional link anyways. Then again only the router side has to handle multiple WANs, whereas the servers job is mainly to reorder packets and blow them into the web. (Is that a correct assumption so far?)

With that setup I reached 250/110 linkspeeds, which is good, but not as good as what I achieved with my openMPTCPRouter setup.

I already tried out settings for redundancy for another usecase, where I want the uplink to be “unbreakable” for live streaming purposes (with WAN Smoothing on and higher values for FEC), and this works as expected.
But this is of course optimized for high availability rather than high speeds.

I would now like to know how to configure the SpeedFusion VPN to have the maximum possible throughput for my other usecase. Do any of you guys have experience with that kind of twiddleing? What can I do to improve actual throughput?

What I can observe is, that the WAN links are not operating at full capacity. Sometimes they are, sometimes they are not. Of course 5G and other mobile connections have a lot of variance in their actual throughput, but I expected my landline to be utilized fully at any time, but this is not the case.

The image shows my 3 WAN connections during a speed test while uploading. The uppermost being my landline, and as you can see, It is not really utilized while the most traffic is transported by one of the 5G links.

I know of TCP congestion and “starving” of TCP links when packet loss becomes too big or when latency rises. (that’s why I activated FEC)
I also know that speedtests are somewhat theoretical and cannot be fully compared with actual real life workflows and data transfers. Still, with my openMPTCPRouter I achieved better speeds, in speed tests aswell as with real life data transfers.
I tested with speedtest.net, fast.com, speed.cloudflare.com and the inbuilt speedtest from peplink within the speedfusion VPN details window, all with the same results, so at least I would rule out the speed test itself as the issue. When using iperf I receive more stable results, but still not the same as with openMPTCPRouter

I experimented with the other settings aswell (Congestion Latency Level, Ignore Packet Loss Event, Disable Bufferbloat Handling, Disable TCP ACK Optimization and buffer sizes).

A bit of insights and knowledge input would be greatly appreciated!

Thanks in advance!
Julian

[WillJones]

I’d start with getting data from the detailed graphs on the speed fusion profile so you can see the traffic distribution over time, with loss/latency also shown.

Like you’ve already done I would fire up the internal SpeedFusion bandwidth test tool and run it for a few minutes so you get a good sample of data, you can also use the WAN analysis tool to prove the bandwidth between the Balance and the FusionHub outside of the tunnel in a repeatable fashion.

One tip you might not know is a lot of the blue question marks in the UI hide advanced buttons behind them, there are some relating to latency cutoff that can be quite impactful, especially in getting SF to saturate a link but not to drive it up to the point of inducing loss.

DWB vs traditional bonding will be a bit more likely to back off a link in my experience, but generally will result in higher overall throughput or less loss/retransmits over time.

Inserting a receive buffer can help a lot too if you have significant latency differences between the various paths though that will set the tunnels minimum latency to whatever value you enter there but for some context I’ve gone as high as a second or so for live streaming where we were using 4x 5G uplinks and it did a lot to bring the tunnel into a very consistent and stable throughout state.

All that said I’ve recently been testing SpeedFusion vs Bondix and have found the latter seems to have lower overheads vs SpeedFusion and appears to perform a bit better under ideal conditions, but when you have really ropey and variable connectivity SpeedFusion has a lot more knobs to tweak and ways to handle that so each has strengths and weaknesses beyond just outright throughput.

Another neat thing with SpeedFusion is you can bond asymmetrically - i.e. only use some connections for upload by default, I’ve had really good results in the past with that too where we rely on say a couple of Starlink dishes to do the heavy lifting for downlink traffic but add in cellular just for uplink which can be really handy especially if you’re mostly streaming out and don’t need to worry too much about download speeds.