Octet

Why did Cisco add EIGRP wide metrics, and how do the numbers change?

Show the answer

Say this first

The classic metric runs out of resolution on fast links. Bandwidth is 10^7 divided by kbps, so a 10 Gbps link already scores 1 and faster links cannot be told apart, and delay is counted in tens of microseconds with the same 10 usec default from gigabit upward. Wide metrics, in named mode, use a 64-bit composite with throughput and latency scaled by 65536 and delay kept in picoseconds. The topology table shows the 64-bit value; the routing table shows it divided by the RIB scale, 128 by default.

Go deeper

The two main terms

  • Throughput = 65536 x 10^7 / slowest bandwidth in kbps
  • Latency = 65536 x total delay in picoseconds / 10^6 With default K values (K1=1, K3=1, and K2, K4, K5, K6 all 0) the composite is throughput plus latency.

Worked numbers for gigabit

  • Throughput: 65536 x 10^7 / 1000000 = 655360
  • Latency for one gigabit interface (10 usec = 10^7 ps): 65536 x 10^7 / 10^6 = 655360
  • Connected gigabit route: 655360 + 655360 = 1310720, shown in the RIB as 1310720 / 128 = 10240
  • Gigabit LAN one gigabit hop away: 655360 + 2 x 655360 = 1966080, RIB value 15360
Output
R1# show eigrp address-family ipv4 topology
EIGRP-IPv4 VR(CAMPUS) Topology Table for AS(100)/ID(10.255.0.1)
Codes: P - Passive, A - Active, U - Update, Q - Query, R - Reply,
       r - reply Status, s - sia Status

P 10.2.0.0/24, 1 successors, FD is 1966080
        via 10.0.12.2 (1966080/1310720), GigabitEthernet0/0/0
Output
R1# show ip route eigrp | include 10.2.0.0
D        10.2.0.0/24 [90/15360] via 10.0.12.2, 00:05:02, GigabitEthernet0/0/0

K6

Named mode adds K6 for extended attributes such as jitter and energy. It defaults to 0 and I have not seen it used in production.

Traps

  • Reading the topology table FD and the routing table metric as different paths. In named mode they differ by the RIB scale.
How did you do?Book 3.4 Routing