Comparing policy-map commands: bandwidth, police, and shape

From: ccie2be (ccie2be@nyc.rr.com)
Date: Thu Dec 16 2004 - 15:37:10 GMT-3


Hi guys,

This is a summary of what these commands do under different conditions. I'm
posting this to
help others and to confirm my understanding is 100% correct. If I state
anything that's misleading or incorrect, hopefully,
someone will point that out.

Here's the scenario:

R1 is connected to an ISP that's policing traffic at 128kbps. 32kbps of
traffic is considered mission-critical and should always get priority
treatment.

R1's config

class-map MISSION-CRITICAL
match access-group name MISSION-CRITICAL

class-map NON-CRITICAL
match access-group NON-CRITICAL

policy-map COMPARE
  class MISSION-CRITICAL
    priority 32
  class NON-CRITICAL
    <action 1, 2 or 3>

service-policy output COMPARE

Action 1 = bandwidth 64
Action 2 = shape 64
Action 3 = police 64

Since the ISP is policing at 128 and by default, 25% of the link's bandwidth
is reserved for the class class-default traffic, the total that the other
classes can reserve and use is 96k.

Behavior when MISSION-CRITICAL traffic is using 32k continuously.

For bandwidth 64, NON-CRITICAL traffic over 64k is dropped.

For shape 64, traffic in excess of 64k is held in shaping queue unless the
shaping fills-up, in which case, it's dropped.

For police 64, traffic in excess of 64k is dropped regardless of what action
is specified for exceed or violate. Even if the action for traffic exceeding
contract is to remark dscp and transmit, since there's no other available
bandwidth, the traffic will be dropped.

Behavior when the MISSION-CRITICAL traffic is NOT using the 32k reserved for
it.

For bandwidth 64, NON-CRITICAL traffic over 64k can "overflow" and use the
unused bandwidth reserved for MISSION-CRITICAL traffic.

For shape 64, the behavior doesn't change from before ie NON-CRITICAL can NOT
"overflow" and use any of the unused priority bandwidth.

For police 64, NON-CRITICAL traffic can burst over 64k up to Be and thus
overflow and use some of the unused priority bandwidth.

I assume the bandwidth reserved for the class-default traffic can NOT be
encroached upon but I'm not 100% sure about that.

If anyone disagrees with anything I've said, please let me know. Thanks

Tim



This archive was generated by hypermail 2.1.4 : Mon Jan 03 2005 - 10:31:27 GMT-3