I have a setup between two offices which utilise 4x connections in one and 3x connections in the other.
What I have found is that when all the connections are running, one wan to wan connection may have high latency and this bring the performance of the entire link down. There is an option to turn off the connection temporarily in the speedfusion wan status menu however this only lasts for 15 minutes.
When I turn off the underperforming links I still have the redundancy that I need but significantly improved performance as the high latency connections have been removed.
Is it possible to keep this setting permanent? High lighted with red line
I’m surprised that a ~50ms difference is enough to cause significant degradation in the PepVPN performance, but then you don’t state what kind of links these are or the capacity of them or the expected performance.
As a way to mitigate this you could try using the latency cutoff value on the PepVPN profiles, this is an option that you can tune per WAN link hidden behind a one of the blue question marks.
Depending on your use case and the type of traffic involved it may also be worth looking at whether using WAN smoothing or FEC might help make these issues less noticeable, whilst both will reduce the overall potential throughput of the tunnel they will potentially help to incase the reliability of the data transmitted over it.
Another option could be to try a different bonding algorithm, although it’s intended for use with cellular links I’ve had good results using the dynamic weighted bonding in similar scenarios as what you describe.
Edit:
Just noticed in your screenshots there is also evidence of packet loss on all the WANs at some stage - is that likely congestion on the WAN connection itself or on the transit path between the two sites? If the former I’d look into whether you can remedy that and if not then DWB and/or a combination of smoothing and FEC could certainly be worth looking into.
Thanks Will, going to give the cut-off latency a try this evening along with firmware update as currently on 8.1.3 which doesn’t have the additional bonding algorithm.
It is rather strange the latency has an overall affect on the link, in testing I was getting very choppy youtube stream over RDP with all wan connections on, with only the lowest latency connections active the stream was significantly smoother.
The packet loss will be due transit path, as its connecting between the UK and India. We are using leased lines as well as a SLA’d P2P line through COLT (ignore the image which shows BT Global). The peplink has been running solidly for some years now however there have been issues with undersea cables recently which directly affects our route between UK and India we have noticed this behaviour.
WAN smoothing is turned on, FEC caused performance issues so is switched off.
You can try DWB from the support.cgi page on 8.1.3 - instead of hitting /manga/index.cgi change to support.cgi and click the “PepVPN traffic distribution“ button. I’d maybe stick with 8.1.3 for now as it’s been a very solid firmware for us in many places.
If neither of those does the trick I’d probably just lower the priority of the troublesome WANs on the PepVPN profile as sounds like it’s maybe better to not use them unless in a failover scenario if they are dragging down the performance that much.
So carried out a fair few tests including DWB bonding algorithm, this definitely was an improvement on the standard bonding but still not perfect. What I found is that this profile used a lot more bandwidth for single stream of a youtube video which usually is around 15-30mbps jumped up to 60-79mbps. I wasn’t keen on running this level of bandwidth utilisation during a working day where many more users are online.
DWB Bonding profile WAN utilisation with 1 user (everyone else offline) streaming youtube video over RDP:
Ended up doing the firmware upgrade to 8.2.0, reverting to standard bonding and modifying WAN mapping instead of using latency cut off as this could cripple the connection if the other lines ever have a latency spike which isn’t uncommon given the distance between UK and India:
I tested DWB bonding with this setup and again there was a significant spike in bandwidth utilisation so stuck to normal bonding and smoothing set to medium (this was my default before). Happy with the performance and everything has been running perfectly today.
