What is a "subnet mask"?

I'm going to use phone numbers as an analogy here. Like any it's an imperfect analogy, but I think it'll help make the point.

In North America, phone numbers are 10 digits long and consist of three parts:

• The area code of three digits. Originally an area code defined exactly that - a geographical area. Area code 206 for example was once all of western Washington state.

• The exchange of three digits. The exchange identified the switching equipment that typically covered a sub-region with in the area code. For example, 788 represents the "Duvall" exchange - located near the city of Duvall in western Washington is a small utility building that houses the equipment that is the 788 exchange.

• And finally, the four digit line number. It's this number that identifies each pair of wires the leave the exchange and arrive at a real telephone set.

So what we take for granted as a 10 digit "phone number" is really a construction of three distinct numbers, each with a specific meaning. (And for the record, many of those meanings have become blurred over the years, particularly with the advent of cellular phones.)

Now let's look at the internet. You already know that each computer connected to the internet has its own equivalent of a phone number, called its "address". Names, such as "www.microsoft.com" actually map to these numeric addresses, such as "207.46.156.220". As you might already be thinking each internet (or IP - for Internet Protocol) address breaks down into components not unlike the 10 digit phone number. The difference is that the components are not always the same size.

IP addresses broken down into what are called "classes". Classes are kind of like area codes; they define where, at the highest level, an address lives. To grossly over-simplify, classes are really just a contiguous block of internet addresses. Within each block though it's still desirable to "subdivide" that network into smaller logical groups. For example a class-A network can have 16 million addresses. It's not a good idea to have traffic for 16 million computers traveling across the entire network. That's where "subnetting" and the subnet mask comes in. The internet protocol's rough equivalent to the telephone's exchange number.

The "problem"with subnets is that each network administrator who's been assigned a range of IP addresses is free to create their own subnets and to define how large they are. It's the subnet mask that defines how big a part of the internet address is to be used as the subnet number.

Unfortunately this is also where we have to start thinking like computers in binary. We'll use these addresses:

  1) 164.109.28.3   [binary: 10100100 01101101 00011100 00000011]
  2) 164.109.27.233 [binary: 10100100 01101101 00011011 11101001]
  3) 164.109.139.4  [binary: 10100100 01101101 10001011 00000100]

And this subnet mask:

  255.255.128.0 [binary: 11111111 11111111 10000000 00000000]

The mask's binary digits are set to 1 to indicate the positions of an internet address that define the subnet. Conversely, it's set to zero for that portion of the address that defines the specific computer on that subnet.

If we do the masking and converting and other what-have-you, we find that of our example addresses #1 and #2 are on the same subnet: 164.109.0.0 while #3 is on a different subnet: 164.109.128.0.

Which after all this brings us to an even more important question: why should we care?

It's all about routing. When information flows across the internet, it flows through equipment called "routers". Routers look at the IP address the information is destined for and decide on the best way to get it there. A subnet is a quick way to know where the information belongs. For example, a packet from our example #1 to example #2 is on the same subnet so routers can take advantage of that information and not send the packet anywhere else. A packet from #1 to our example #3 however is destined for a computer on a different subnet and the routers handling that packet will know to send it along a different path to get there.

To use our telephone analogy again, if I'm in Duvall making a call on my 788-xxxx telephone to another 788-xxxx telephone then the phone equipment knows that it doesn't have to try anywhere else - it all happens within the Duvall exchange. On the other hand, if I try to call a 483-xxxx number then the 788 exchange needs to route my call to other equipment within my area code that knows how to find the 483 exchange.

As you can imagine, I've only scratched the surface here. IP addresses, subnets and supernets, routing, special purpose addresses, network address translation; the internet's a complicated world. If you're interested in reading further here are some resources:

• What is an IP address? over at How Stuff Works.

• How Routers Work, also over at How Stuff Works.

• Understanding IP Addressing in the Webopedia.