CFWorx

VLSM: variable length subnet masking

· in Cisco U

The problem with one mask

Plain subnetting means one mask for the whole network, so every subnet is the same size. Real networks need a mix: a user LAN might need 50 hosts, a point-to-point WAN link only needs 2.

Example: 172.16.0.0/16 cut into /24s gives 256 subnets of 254 hosts. Put a /24 on a WAN link and 252 addresses die there. A LAN with 19 hosts doesn’t need 254 addresses either; a /27 (30 usable) fits far better.

What VLSM does

Use different masks inside the same network, sized per subnet. In practice that means taking an existing subnet and subnetting it again: “subnetting a subnet.”

The rule of thumb: allocate the biggest subnets first, then carve smaller ones out of what’s left.

The math is the same as normal subnetting, see borrowing bits and subnet increments. You just apply it more than once.

Worked example: 172.16.32.0/20

Start: 172.16.0.0/16 subnetted with a /20 (4 bits borrowed) gives 16 subnets of 4094 hosts each. One of them, 172.16.32.0/20, gets assigned to a region needing several 50-host LANs plus WAN links.

LAN subnets

50 hosts needs 6 host bits (2^6 - 2 = 62 usable), so the mask is /26. Borrowing 6 more bits from the /20 yields 64 subnets.

console
Subnet             Hosts          Broadcast
172.16.32.0/26     .1  to .62     172.16.32.63
172.16.32.64/26    .65 to .126    172.16.32.127
172.16.32.128/26   .129 to .190   172.16.32.191
172.16.32.192/26   .193 to .254   172.16.32.255

WAN subnets

Grab one unused /26 (172.16.33.0/26) and split it with a /30. That’s 4 more bits borrowed, giving 16 subnets of 2 hosts, exactly what a point-to-point link needs.

console
/30 subnet      Hosts         Broadcast
172.16.33.0     .1  to .2     172.16.33.3
172.16.33.4     .5  to .6     172.16.33.7
172.16.33.8     .9  to .10    172.16.33.11
172.16.33.12    .13 to .14    172.16.33.15

The increment for a /30 is 4, and the broadcast is always one less than the next subnet address.

Crossing an octet boundary counting backwards: the network before 172.16.33.0 ends at 172.16.32.255. Think dollars to pennies, where 255 plays the role of 99.

Route summarization

The second big win from VLSM’s hierarchical layout: many child subnets can be advertised as one parent route (route aggregation). 172.16.14.0/24, say, covers everything carved out of it, /27 LANs and /30 WAN links alike.

Fewer routes in the table means faster lookups, less CPU, smaller updates.

Protocol support

VLSM needs classless routing protocols, the ones that carry the mask in their updates: RIPv2, OSPF, EIGRP.

Classful protocols (RIPv1, IGRP) omit the mask, so they can’t do VLSM. Both are dead tech, listed for exam trivia only.

The lecture tangents from this session grew into their own posts: