Subnetting by hand, still

Every network engineer learns to subnet and then most of them forget, because there is always a calculator nearby. I kept doing it by hand, mostly out of stubbornness, and it turned out to be one of the more useful habits I carried into SRE work.

The one trick

A /24 is 256 addresses. Every bit you borrow halves that. So a /26 is 64 addresses, a /27 is 32, a /28 is 16. The block size is the only number you need; networks start at multiples of it.

Given 10.20.30.77/27:

• block size is 32;
• 77 falls in the block starting at 64 (64, 96, 128, …);
• network 10.20.30.64, broadcast 10.20.30.95, hosts .65–.94.

That is the whole method for anything /24 and longer. For shorter prefixes the same arithmetic applies to the third octet: a /22 has a block size of 4 in the third octet, so 10.20.30.77/22 lives in 10.20.28.0–10.20.31.255.

Why it matters when reading routes

Longest-prefix match is a comparison of block sizes. When you can see at a glance that 10.20.30.0/23 covers 10.20.31.77 and 10.20.30.0/24 does not, you read a routing table the way a native speaker reads a sentence instead of translating each word.

IPv6 is easier, not harder

Nobody subnets IPv6 below a nibble boundary in practice, so the arithmetic is in hexadecimal digits: a /48 has 16 bits of subnet space, which is four hex digits, which is 65,536 /64s. 2001:db8:abcd:1234::/64 is the 0x1234th /64 in the 2001:db8:abcd::/48. There is no broadcast address and host counting is meaningless, which removes half of the IPv4 trivia.

The subnet calculator on this site shows the binary view next to the answer for exactly this reason: the point is to make the block boundaries visible, not to hide them.