MikroTik

Hi
I cant remember how the correct Therm is for that. Some Company have an mode where 1 Packet is send truth (some/) all WAN ISP and the Packed who arrived first will be taken and the other duplicated Packed get tossed away. That repeat for every single other one to. Image this there is a Contract who must be signed by the other Company to and you send 3 Employees with 3 different Airlines to the other Company and the first who arrive at there HQ get the Paper signed and the other 2 get tossed away. So you can be sure that:
a) every Packed arrives on the other End even on the way there a packed is lost
b) the fastest packed get picked so when one (Mobile) Connection is overloaded there is no down time.

Who do I need them? In my Country we have 3 Cellphone Operator with (100%) Unlimited Internet. I want stream some events remotely and everyone an imagine at some Open Air Concert for example it can get very tricky to get reliably (!) Data from A to B.

This is a mikrotik forum bud.......... not applications for fancy routing.

What if all airplanes crash and not reach HQ...

And imagine if top management and all the development gurus were on the crashed airplanes!

Even if you could manage to implement duplicating packets through multiple interfaces, I doubt other side will accept it normally, since it would expect a packet flow from a single endpoint. Otherwise, how they should preserve packet sequence, especially with LTE delays.

The other Side would be the same (Virtual, Software,..) Router as the one who send the Packeds.

What you're talking about is called packet duplication and is an SD-WAN vendor technique. Cisco, Fortinet, Velocloud, and so on all use it.

From an article by Cisco (their product is Cisco Catalyst SD-WAN):

https://learningnetwork.cisco.com/s/art ... r-starlink

Packet duplication is an SD-WAN technique designed to enhance reliability by mitigating packet loss. When a critical packet is scheduled for transmission, the router creates an identical copy. Both the original and duplicate packets are then sent over separate SD-WAN tunnels to the destination. The receiving device detects and discards duplicate packets, ensuring only one copy is processed. This method relies on multiple available tunnel paths and is typically configured for applications sensitive to packet loss, such as voice or video. While improving reliability, packet duplication introduces additional network overhead in terms of bandwidth and processing resources. It is configurable through SD-WAN management platforms to specify target traffic and redundancy levels.

Here's a 4-year-old Reddit post where a few people share their experiences:

https://www.reddit.com/r/networking/com ... in_sdwans/

Apparently ZeroTier does Multipath. The question remains as to whether or not MikroTik's implementation supports it yet. But on Linux, you have a few options:

Standard policies

• active-backup: Use only one primary link at a time and failover to another designated link.
• broadcast: Duplicate traffic across all available links at all times.
• balance-rr: Stripe packets across multiple links (not for use with TCP.)
• balance-xor: Hash flows to specific links.
• balance-aware: Auto-balance flows across links.

(from https://docs.zerotier.com/multipath/)

Interesting!! Be cool for the sirbryan to conduct single router to single router tests of this tech, comparing zerotier to wireguard performance.........

This is a mikrotik forum bud.......... not applications for fancy routing.

maybe OP wants this -> viewtopic.php?p=1105740 (and guess who is OP of this )

OK, so let's take it more seriously.

The broadcast mode of bonding indeed works well as for multiplication of the packets, but it does nothing at all regarding "tossing the late ones". The reason is that while broadcasting, the sending end does not add any information to the packets that would allow the receiving end to distinguish duplicates created by a broadcasting bond from duplicates created "naturally" - as an example, think of ARP requests that contain no information that would distinguish them from one another.

So rather than broadcasting the payload itself, you would need to broadcast transport packets of some tunneling protocol encapsulating the payload ones, which would be equipped with some sequence numbers allowing the receiving end to drop the late copies. I hazily remember someone here on the forum already did some research - I would assume IPsec-AH should take care of this but I may be wrong.

But that's not all yet - as bonding can only bond together L2 interfaces, you need to create L2 tunnels to carry the bonded packets; you can choose between EoIP, VxLAN, and L2TP in bridging mode. As neither EoIP nor VxLAN can traverse NAT on their own, they need to be further encapsulated into L2TP, bare IPsec, or Wireguard, hence using L2TP in bridge mode directly seems the best way to me.

So the summary is that you end up with at least two layers of tunneling if you want to go this way, but what is worse, even with three operators you cannot be sure your traffic will get delivered with a consumer grade mobile contract.

Therefore, if your streaming is meant to be a serious business, I'd say you should talk to the mobile operators and ask them for SIMs with prioritized handling. The traffic to/from such a SIM will get priority even in a crowded cell, but of course such a SIM is provided under a more expensive contract.

i think this is a very interesting topic, Once again, FEC support for tunnels to remote sites resurfaces in the conversation

Interesting!! Be cool for the sirbryan to conduct single router to single router tests of this tech, comparing zerotier to wireguard performance.........

According to the various Reddit posts etc. that I came across when researching my original reply, Wireguard is faster than ZeroTier. And in my own experiments I've found Wireguard converges pretty quickly if you have traffic flowing and the OS is quick to reroute traffic (i.e. during interface down events--plug, unplug, plug, unplug Starlink...).

In my experiments, the primary connection is my own network, and a Wireguard tunnel over a different provider is the second link. I have OSPF + BGP established over both, and can use BFD to detect pretty quickly if the primary link is down. This works and works rather well with both Starlink and cable; I have it in production at two locations.

The difference is it only uses one connection at a time and doesn't offer any kind of FEC or packet duplication. I can set the OSPF costs to be equal, and let ECMP handle load-sharing (which also works). But then some flows are randomly limited to the worst bandwidth in one direction (cable and Starlink upload is 20-30Mbps), and the latency changes depending on whether the traffic stays on-net or goes via the Wireguard link. So in the end, the backup connection sits idle most of the time.

I think Amm0's response in the linked thread does a good job explaining L3 (OSPF+ECMP+BFD) load-sharing vs. L2 (EOIP tunnels) with some kind of LACP or broadcast bonding. I will have to test the latter, however. That would be interesting.

Okay either I have to travel to UTAH or we get together on (take your pic skype/discord/teams) to discuss those very configs...........
Use case: seamless failover between two WANS on the ground site, using CHR as the public IP hitting the internet.
EOIP within wireguard.

Trying to do better than check-gateway=ping OR netwatch.

Check my link in my profile. There's a way to contact me...