IPv6 Subnet Calculator

An IPv6 address is 128 bits, split by convention into a 64-bit network prefix and a 64-bit interface identifier. That lower half is what Stateless Address Autoconfiguration and RFC 4941 privacy addresses use to generate addresses, so shortening it breaks them. The result is that essentially every ordinary LAN gets a /64 regardless of how many devices sit on it, and you size networks by counting /64s rather than by counting hosts.

65,536 /64 subnets. Each bit between the site prefix and the /64 boundary doubles the count, and a /48 leaves 16 such bits, so 2^16 = 65,536. That is why a /48 is the common allocation for an entire site or organisation: a /56, frequently handed to residential customers, leaves 8 bits and therefore 256 /64s, which is still far more than a typical home needs.

No. IPv6 has no broadcast address at all, replacing it with multicast, and it does not set aside a network address the way IPv4 does. The subnet-router anycast address, which is the all-zeros interface identifier, is reserved on a /64, but you do not lose the two addresses per subnet that IPv4 costs you. In practice the usable count on a /64 is so large that these reservations never affect planning.

Drop the leading zeros within each of the eight hextets, then replace the single longest run of consecutive all-zero hextets with a double colon. The critical constraint is that :: may appear only once in an address, because more than one would make the number of zeros it stands for ambiguous. So 2001:0db8:0000:0000:0000:ff00:0042:8329 compresses to 2001:db8::ff00:42:8329.

It helps considerably. A nibble is four bits and maps to exactly one hexadecimal digit, so prefixes at /48, /52, /56, /60 and /64 fall on digit boundaries and keep the subnet number readable at a glance. Subnetting at a length like /59 works technically but splits a hex digit across the prefix boundary, which makes addresses harder to read and reverse-DNS delegation more awkward than it needs to be.