Use Case - SpeedFusion Dynamic Weighted Bonding

Could I sanity check my troubleshooting process with you for a client deployment please?

We have a Transit Duo deployed in a Farmhouse in a remote location in the UK.

Cell1 had Vodafone attached to omni (Puck-2) to connect to potentially 2 or 3 different masts along the a main road near the Farmhouse.

Cell2 has EE which is directional to a mast on another main road. This had superb throughput in testing but has the potential to become congested during certain periods of the year, when tourists are in the area.

House has a poor copper ADSL of 2/0.4Mbps connection and no near-term chance of fibre, but internal infrastructure of Netgear 24-port switch and Ubiquity Aps was already present.

At handover (before Christmas) we tested Wi-Fi & Ethernet ports at least 60/20Mbps with at router results of 110/33.

Initially feedback was good and large volumes of data were being used (200GB on 25/12) half of which was from a single device (xbox).

However since then the owner has reported issues with sky / iplayer buffering / dropped zoom and even dropped radio streams.

I looked at the IC2 WAN quality charts and couldn’t identify any particular concerns, but implemented some SFC bonding for Zoom/Teams and also added Sky Q box but seems not to have made much difference.

So I have downloaded the WAN quality CSVs for January and focused on latency and created the following summary:

EE

Friday 13th, 21:45 – 21:58 there were 9 samples that had 1-second or greater latency 

To-date in January EE has had total of 19 samples with >1s, these were spread over 5 days; those 15-mins on the 13th being the worst period.

Total for the month so far out of 340498 samples: 

BAD or no latency reported = 104,  (33 of the no reported latency comes from the weekly reboot that is scheduled for 04:30 every Thursday)

POOR = 317, 

GOOD = 3330, 

Excellent = 336745  (98.9%)

Vodafone

Monday 22nd ,16:14-16:19 & 17:43-17:44 there were 14 samples of 1 second or more latency

Total for the month so far out of 260110 samples: 

BAD or no latency reported = 83, 

POOR = 248, 

GOOD = 2361, 

Excellent = 257417  (98.9%)

Where I defined latency

Bad is >1 second
poor 0.3-1.0s
good latency 0.3s and
 excellent <0.1s

Which again don’t indicate any issues to me…

A 120s SpeedFusion TCP download test just now gave the following results:

There was certainly some variation in the speeds reporting but majority were in the 40-90Mbps with 2 sub 10Mbps and 2 over 195Mbps – I confess to not seeing enough of these to know if this is a good result but with 4 streams each achieving 6-10mbps it looks like it should support their family home needs…?

Is there anything else on the Peplink side of things I should be checking before turning my attention to the client’s network infrastructure?

Response:-

I would agree with your assumption about bad / poor / good / excellent latency levels.

Have you tried adding an WAN smoothing or Forward Error Correction in the SFC tunnel?

If you are using Dynamic Weighted Bonding, I would recommend you try disabling the Bufferbloat handling and set the Packet Jitter Buffer to 0ms:

Congestion latency level, I usually leave that at default, but for some installations where latency is important, I'd change that to "Low" - this will treat the connection as congested at a lower latency level.

If this doesn't improve things, then I would look at the internal WiFi / AP's - with Peplink AP's the minimum threshold can be set, so an AP will drop the client when the threshold goes below a certain level, allowing the client device to connect to a closer AP with better signal strength.

Its possible that the Zoom calls on a Laptop where going over the "wrong AP"...  if the laptop was in the lounge and joined the network with a good signal, when it is moved to the kitchen table (hypothetical scenario) the Laptop is still connected to the AP in the lounge, rather than the AP in the kitchen.