Seamless redundancy in networks

Redundancy with zero downtime, at the cost of double the bandwidth. HSR and PRP provides that for you. All traffic is duplicated to ensure that at least one copy arrives even if a cable breaks. Many other protocols, like spanning trees, requires reconfiguration when the topology changes.

Setup

Let’s start by creating some veth interfaces that we can communicate over. One is put in a network namespace. ARP can be a bit wonky when both ends are in the same namespace.

Let’s start by creating two namespaces and veth ports assigned to them. This requires sudo, or CAP_NET_ADMIN + CAP_SYS_ADMIN capabilities for ip. ARP can be a bit wonky if they are in the same namespace, so this is easiest. Sudo also isn’t required once inside the namespace.

sudo ip netns add ns1
sudo ip netns add ns2
sudo ip link add dev veth1 netns ns1 type veth peer name veth2 netns ns2
sudo ip link add dev veth3 netns ns1 type veth peer name veth4 netns ns2

Open two terminals, and attach in one for each.

sudo ip netns exec ns1 bash

sudo ip netns exec ns2 bash

In namespace ns1 execute the following to set up an HSR interface with an IP address.

ip link set dev veth1 up
ip link set dev veth3 up
ip link add name hsr1 type hsr slave1 veth1 slave2 veth3 version 1
ip addr add dev hsr1 198.18.2.1/24
ip link set dev hsr1 up

And the same for ns2.

ip link set dev veth2 up
ip link set dev veth4 up
ip link add name hsr2 type hsr slave1 veth2 slave2 veth4 version 1
ip addr add dev hsr2 198.18.2.2/24
ip link set dev hsr2 up

If you instead wish to use PRP, add proto 1 at the end when creating the HSR interface. They use the same underlying principles.

The topology we have now looks like this.

    ns1               ns2
 198.18.2.1        198.18.2.2
+----------+      +----------+
|          |      |          |
|  +-veth1-+------+-veth2-+  |
|  |       |      |       |  |
| hsr1     |      |     hsr2 |
|  |       |      |       |  |
|  +-veth3-+------+-veth4-+  |
|          |      |          |
+----------+      +----------+

Testing the redundancy

The hsr1 interface owns the ports veth1 and veth3. Any traffic sent on hsr1 will be duplicated and sent across both ports. Any packets received has the first copy delivered upwards, and the second (if there is one) is discarded. Pinging from ns1 should now reach the IP of hsr2 and get a reply.

ping 198.18.2.2

Break one link and see that the ping keeps going.

ip link set dev veth2 down

And bring it back up again.

ip link set dev veth2 up

Try the same with veth4. Try setting both veth2 and veth4 down at the same time, now the pinging will stop.

Theory of Operation

I already mentioned the basic concept, duplicate traffic. If one packet doesn’t go all the way, hopefully the other does. And hopefully someone can get out and fix the broken cable before something else breaks. There are ways of combining HSR and PRP to get more levels of redudancy, but that is out of scope for this post.

HSR, High-availability Seamless Redundancy, is a ring protocol. Every device in the ring must support HSR. A header tag is attached at the start of the ethernet packet. Most important is the sequence number, a 16-bit integer counter for each source MAC address the interface transmits. The counter increments for every packet sent from the corresponding source MAC. This means duplicates are identified by their source MAC + sequence number. If the receiver sees two matching packets within a short timespan, it assumes the second one is a duplicate and discards it. The counter wraps around and numbers can be reused.

Because HSR uses a header tag it requires every device in the redundant path to support it. PRP, Parallel Redundancy Protocol, does away with that and instead uses a trailer, a very similar tag that goes at the end of a packet. The trailer contains a small checksum, to somewhat guarantee that the packet it sees is a PRP packet, and not just part of a normal payload. PRP is not a ring protocol. The two ports should instead go to a separate LANs, with the receivers also having one port to each LAN.

Because the trailer makes it look like a normal packet, any kind of switch can be used in between. It can be combined with RSTP within the LAN to provide redundancy for less critical devices that don’t support the PRP dual connection.

Teardown

Exit the namespaces and run the following.

sudo ip netns del ns1
sudo ip netns del ns2