You might have heard of TCP/IP, or at least IP addresses. The IP part of this acronym stands for Internet Protocol. The Internet Protocol is, besides Ethernet, one of the most fundamental protocols of today’s global networks.
Ok, wait a second, does this mean Ethernet, which we previously talked about, is obsolete and IP is the new thing?
Well, not exactly. It is true that Ethernet can only be used between devices on the same network, but IP does not replace it. Instead, IP is transported on top of Ethernet.
Let’s start with the basics: we want to assign each of our devices a unique address. Preferably globally unique so that the address can be used on a global level.
We do this by taking a very large number block, say zero to roughly four billion. We can store this number on 32 bits of space. You may notice that four billion addresses are not even enough for every human on Earth, we’ll get to that later.
Now, we carve up this vast number space and give each provider a chunk of it, say 0-1023 for one network provider, 1024-2047 for the next, and so on.
Each network provider can assign the IP addresses to their devices as they see fit. Early on this would be done by hand, later this would be automated using a protocol called DHCP.
So each device has a unique number. Let’s call it by its proper name: an IP address. How do we get a data packet from a computer in the network of provider A to a computer in the network of provider B?
For our example these two providers must agree to exchange data. They deploy a common networking device called a router that is connected to both networks. Additionally they will feed the router the information on which network can be accessed over which cable. This is called a routing table.
Sending data on a local network
When our computer on network A wants to send some data to another computer within the same network, the situation is simple. It too has a routing table and knows which network block it is on. Say, 0-1023.
Similar to Ethernet frames, IP packets also have source and destination addresses in them. Our computer now writes its own IP address into the source field and the other computers IP address into the destination field.
However, it can’t just simply blindly fire this IP packet onto the local network. It needs to package the IP packet into an ethernet frame. For that it needs the MAC address of the target computer.
The MAC address is obtained using the Address Resolution Protocol (ARP) or the Neighbor Discovery Protocol (NDP), depending on which IP version we use.
Once the MAC address is found out, our computer creates an Ethernet frame with the target MAC address, puts the IP packet inside and sends it off on the local network.
Sending data outside the local network
Now here comes the tricky part. What if the target computer is on network B and not on the local network?
As before, our computer creates the IP packet, but also realizes that the target computer is not on the local network. In its routing table it will find an entry for default gateway: the IP address of the router. The default gateway entry means that any packet that doesn’t have a more specific rule in the routing table, will be sent there.
So, instead of using ARP or NDP to find out the MAC address of the target computer, instead the MAC address of the router is looked up. Our computer creates an Ethernet frame with the router as a destination and puts the IP packet inside.
The router then unpacks the Ethernet frame and looks at the IP packet inside. Using its own routing table it determines where to send the packet next. Since it is directly connected to network B directly, it can do an ARP or NDP lookup on network B to figure out the MAC address of the target computer, pack up the IP packet in a new Ethernet frame and send it out over the wire on network B.
So in essence, there is a separate Ethernet frame for each network the IP packet travels through and each router or device evaluates the IP packet separately.
IP address exhaustion
Now, as I’ve indicated before, around 4 billion addresses are not enough even if just every human on Earth has one device. The older, version 4 of the internet protocol, unfortunately uses 32 bits for addresses, so we are rapidly reaching the point of exhausting all available IP addresses.
The newer, version 6, of the internet protocol uses 128 bit addresses, so every grain of sand in the Sahara could get an address, but the adoption has been slower than expected. The reason for this is partially die to the fact that IPv6 brings in a lot of new features, which makes it a bit of a learning curve to deploy.
To work around IP exhaustion, Network Address Translation (NAT) was invented. Home routers regularly assign devices
IP addresses from one of the so-called private IP ranges (
172.24, etc) and then, when a packet needs to
go into the outside world, change the source IP address of the packet.
They also remember the packet, so if a response is received, it can be sent back to the private IP address it came from. This is also known as masquerading.
Since the IP exhaustion has reached big providers, they sometimes even deploy NAT on a massive scale, commonly referred to as Carrier-grade-NAT or CGN.
I’ve mentioned that IP works on a global scale. As you might imagine, coordinating the routing tables of each and every router in the world is a massive effort.
That’s why this isn’t done manually, but by using so-called routing protocols such as the Border Gateway Protocol or BGP. Using BGP providers announce to each other which routes are available through their connections and the routers automatically update their routing tables.
This, of course, can cause problems, which is why internet providers need to make sure they only accept routing announcements from parties they belong to. But that is a story for another article.