Reference Guide
Configuring the Hash Algorithm Seed
Deterministic ECMP sorts ECMPs in order even though RTM provides them in a random order. However, the hash
algorithm uses as a seed the lower 12 bits of the chassis MAC, which yields a different hash result for every chassis.
This behavior means that for a given flow, even though the prefixes are sorted, two unrelated chassis can select
different hops.
FTOS provides a command line interface (CLI)-based solution for modifying the hash seed to ensure that on each
configured system, the ECMP selection is same. When configured, the same seed is set for ECMP, LAG, and NH, and is
used for incoming traffic only.
NOTE: While the seed is stored separately on each port-pipe, the same seed is used across all CAMs.
NOTE: You cannot separate LAG and ECMP, but you can use different algorithms across the chassis with the same
seed. If LAG member ports span multiple port-pipes and line cards, set the seed to the same value on each port-
pipe to achieve deterministic behavior.
NOTE: If you remove the hash algorithm configuration, the hash seed does not return to the original factory default
setting.
To configure the hash algorithm seed, use the following command.
• Specify the hash algorithm seed.
CONFIGURATION mode.
hash-algorithm seed value [linecard number] [port-set number]
The range is from 0 to 4095.
Link Bundle Monitoring
Link bundle monitoring is supported on the S4820T platform.
The link bundle monitoring feature allows the system to monitor the use of multiple links for an uneven distribution. A
global default threshold of 60% is the usage percentage for the bundle; when the system reaches this threshold, it
begins monitoring the configured ECMP groups for uneven distribution. Links are monitored in 15-second intervals for
three consecutive instances. Any deviation exceeding 10% among any of the bundle links sends a syslog and an alarm
event is generated; for example, 01:16:25: %STKUNIT0-M:CP %IFMGR-5-
BUNDLE_UNEVEN_DISTRIBUTION: Found uneven distribution in ECMP-GROUP bundle 1.
When the deviation clears, another syslog is sent and a clear alarm event is generated; for example, 01:35:14:
%STKUNIT0-M:CP %IFMGR-5-BUNDLE_UNEVEN_DISTRIBUTION_ALARM_CLEAR: Uneven
distribution in ECMP-GROUP bundle 1 got cleared.
The link bundle utilization is calculated as the total bandwidth of all links divided by the total bytes-per-second of all
links, as shown in the following example.
Example of Viewing Link Bundle Monitoring
FTOS# show link-bundle-distribution ecmp-group 1
Link-bundle trigger threshold - 60
ECMP bundle - 1 Utilization[In Percent] - 44 Alarm State - Active
Interface Line Protocol Utilization[In Percent]
Te 0/0 Up 36
Te 0/1 Up 52
294










