So for many, many years, I've used various types of hardware that supports multiple WAN's in a kind of failover/failback mode. As they got more advanced throughout the years, some of them supported aggregation, but only in the sense of multiple clients (ie users) would be aggregated seemingly well, which is not a useful scenario for me as a single person.

The idea of VPN bonding is that it can achieve single thread aggregation (single connection, like a single HTTP download request as an example) over two uplinks (or more) concurrently, while also providing all the usual redundancy.

Because of the way the connections are bonded, there is zero delay in failover. With a traditional failover, there would be about a 10 second delay as the router figures out one of the uplinks died, then starts pushing everything solely over the other link, but this delay is exacerbated because on your Windows workstation (ie Sierra, YouTube, webinars, whatever) the connection (for instance, HTTP) would be broken, then a re-connect needed to be performed by the application. This is not seamless, and while for less demanding purposes it is fine, it's not acceptable for a real professional environment.

For my test I'll be using Zeroshell, a open source platform that I'll install on a NUC type device (overkill honestly) between my switches and my routers. The Zeroshell device will become the new gateway for the LAN, and it will then handle the uplink traffic to the actual routers leaving my site (ie Microwave and Fiber).

To make this type of VPN bonding work, you also need a pair of ports on a remote server that has more bandwidth than the combined total of your two (or more) aggregated ports on the local site. I do, in Chicago. I already use one of my servers in Chicago as a VPN, although there is some concern here with this type of bonding and latency, but I am hoping it won't dramatically reduce efficiency.

So on the server side in Chicago will also set a Zeroshell installation, just a VM running under a physical box on Debian. It will have two dedicated IP's assigned to it (the VM), one for each virtual interface.

On my local site, I'll also have two IP's, one for Microwave and one for Fiber. These will be bonded and connect to the bonded interface in Chicago and form a LAN-to-LAN VPN (Layer 2).

The advantage of this type of setup is that there is no client-side configuration necessary (ie: Windows boxes, Linux boxes, PS4, etc) on my local network. They all just continue working like normal. No special configuration necessary. The aggregation and VPN takes place transparently on the Zeroshell router itself (instead of for example using OpenVPN on your workstation).

Wish me luck. I'm not sure when I'll have time to make some progress, but it's on my list. Higher priority recently because my primary internet connection (Microwave) has been real flaky recently with as many as 20 disconnects per day, each lasting only about 2 seconds, but causing constant reconnects in Sierra Chart and making it impossible for me to put on new webinars.

Mike

I should mention there are commercial hardware solutions that do this, they seem to cost about $2,000 per side (so around $4,000 total, at least) and aren't doing anything different on the inside than what I am proposing. My solution costs virtually $0, you could run it on a Raspberry Pi for $35 if you wanted.

In either case, you need server hosting for the datacenter end with the high bandwidth. Still, you can find solutions for that for under $100 a month, maybe in some cases with questionable networks as little as $20 a month.

I'm already paying almost $1,000 a month for my co-lo hosting in Chicago for all my various servers. I own all this hardware, the price is for co-lo only for the quality networks, not for hardware leases.

Mike

Here is an alternative if you are looking to only use bonding on a per-machine level (like your primary workstation). It is a software based VPN that supports channel bonding, $20 a month software lease with a maximum bandwidth of 50 Mbps.

Channel Bonding - Speedify

And this is a good explanation of what bonding is (regardless of Speedify or just using normal VPN bonding on hardware):

Quoting 

Setting up round robin via managing routes will spread your sockets evenly between all of your Internet adapters. If one fails, its sockets are all broken, and the next sockets will be created on still working Internet connections. There are a couple problems with this approach: 1) sockets are divided evenly between adapters even if the adapters are not the same speed, 2) If an adapter gets disconnected all of its sockets are broken, and 3) there's no relationship between sockets and traffic, some sockets send 1KB and some send gigabytes, and you won't know till afterwords. Most obviously this means that round robin is of no help at all with streaming video: Netflix for example, will send you 3.5 GB over a single socket while watching a movie. Speedify addresses those issues with its channel bonding VPN approach. It doesn't work on sockets, it works on the packets that they're made of, so it can will take a single socket and split it across multiple adapters. This is important for things like those Netflix streaming videos. It can also actually move live sockets off of failed adapters and onto the still working ones without breaking them.

Mike