Good Day,
I'm updating a QoS policy for a bunch of tunnel interfaces that we run over our WAN service. Each tunnel connects to a remote site which has a 5M internet service. Below is what I have as a standard policy that I apply facing the site (Download) on our core router. It's purpose is to accomplish:
1) LLQ for voice traffic
2) somewhat protect (minimum bandwidth during congestion) Collaboration Meetings, Control traffic and Business traffic, which is identified by an ACL and qos markings
3) everything else goes in class class-default
config:
policy-map std_child
class Collab_Voice
priority 1000
class Collab_Cntrl
bandwidth percent 4
queue-limit 256 packets
class Collab_Meetings
bandwidth percent 50
queue-limit 256 packets
class Business
bandwidth percent 18
queue-limit 256 packets
class class-default
queue-limit 256 packets
fair-queue
random-detect
!
policy-map std_parent
class class-default
shape average 4500000
service-policy std_child
I shape to 4.5M to allow for overhead and to ensure I don't hit the strict 5M policer the local ISPs have inplace. 1) does this seem like a valid policy (any adjustments?) and 2) running a speedtest from a remote site as a test, I noticed that traffic in the Business class was getting someone 'starved' even though the speedtest traffic is in the class class-default. there was no evidence of drops on increasing Q counters in the Business class during the test. I understand that other than the priority class, the bandwidth specified is not reserved, it is just the minimum available during congestion (caused by hitting the shaper).. my confusion lies within the area of, outside of the LLQ, during congestion, how are the other Collab and Business classes serviced when something appears in the default class that wants all of the bandwidth and business traffic is present? are the classes that are specified serviced first from top to bottom?
thanks,
Al
