A network for a home cloud data center - an Internet from several 100Mbps -1Gbps channels from different providers with channel aggregation and redundancy?

Question

network hardware

Situation:
There is an application for which the performance of the server processor is not very important, but it needs fast Internet and a lot.
There is a desire to make a cloud at home and host it at home, ensuring the fault tolerance of everything. Home Internet costs 6 times cheaper than the Internet from the data center, and in general, it’s more pleasant to have servers nearby, and not somewhere in Germany ...
The application, for example, will live in a Docker container, CoreOS (Linux) is installed on the servers . Let's say 4 servers in the cloud. Each server has a network card for 4 gigabit ports (this is scenario A, scenario B [after upgrading servers in 3-5 years]: for 4 10-gigabit ports).
Let's say, three providers are taken (I think 3 providers will not stop working exactly at 1 moment if there is no emergency in the whole city).
2Com, 2 contracts (2 wires), 1000Mbps each, Ethernet connection, IPoE.
Beeline, 2 contracts (2 wires), 365 Mbit/s each, Ethernet connection, IPoE.
Hello Incognito, 2 contracts (2 wires), 1000 Mbit/s each, connection via PON, Ethernet via optical modem.
Questions:
1.
what two (for fault tolerance) pieces of iron are needed so that there are 3 WAN ports, functionality for link aggregation from the provider (out of three channels to make one at 2365 Mbps) and for link aggregation between this piece of iron and each of the 4 servers (that is, each server will connect to it with 2 channels so that the speed increases to 2 Gigabit / s), that is, 8 LAN ports.
That is, in the case of scenario A, there will be two grids, each with 2365 Mbit / s communication with the outside world, within the network between 2 Gigabit / s devices. In the case of scenario B, the connection with the outside world is the same, but inside the network 20 Gigabit / s.
Tell me for both scenarios, please, what will you need to buy? .. Or, perhaps, it will be cheaper to assemble (based on the same Supermicro) than to buy? .. Ceteris paribus, preference is in favor of Cisco.
Just write, please, which ciscos (I think, as a lover of elite pieces of iron (Apple computers and Dell servers), I should pay attention to the switch market leader) will solve the problem in scenario A, which ones in scenario B.
2. please explain how to correctly do link aggregation between this switch (or a server as a switch) and providers:
a.)https://help.ubuntu.com/community/UbuntuBonding there is some kernel module and apparently balance-alb mode (the best of the three available without interfering with the provider's hardware) b
.) www.shorewall.net/MultiISP.html
.) www.nongnu.org/quagga
d. ) www.multipath-tcp.org
e.) something else
3. Similarly, what is the best technology for link aggregation between this switch and cloud servers?
4. In general terms, I designed a model of a fault-tolerant network above (2 different switches, 3 different providers), is it correct? .. or can / need to improve something? ..
PS I apologize if I wrote some things crookedly, I'm a teapot in building networks, confusing routers with switches, and WAN ports with uplink ports, I've never done this, but I'm ready to figure it out ... At first, I suppose, in the role of pieces of iron there will be ZyXEL Keenetics with OpenWRT firmware, but... this is the most initial solution and I won’t have some of the things described above with it... I’ll also add that I would like to be able to increase both the speed of the Internet and the number of servers in the cloud. ..
I suppose my question and answers to it will be useful to many IT people who, like me, would like to start their own data center at home, stop paying Hetzner and move all their projects home...

Answer 1

A successful startup will rent a stapler

Fault tolerance is achieved not so much by connecting to two "providers", but by connecting to different traffic exchange points. Uninterruptible services
for the server and for "provider" switches are not taken into account.
If the project needs more than 10 Mbps and it does not pay off, it is better to kill it
. the project can come in for tea mask shows, which is quite unpleasant when they also come into your house

Answer 2

The correct, but unrealistic scenario:
1) register an autonomous system (AS) and a /24 grid of PI addresses.
2) we install a Cisco ASR 1k router with a line card for several ports (for uplinks of providers). Or instead of a catalyst 3850 switch card
3) on the ASR we set up BGP peering with each provider
As I understand it, the task comes down to distributing video content. You can consider a simpler option with DNS round robin on the PA addresses of providers.