RADIUS Counters

Per default, the session counters are calculated using the LIF statistics which include all traffic received on the logical interface (InLIF) and all traffic sent on the logical interface (OutLIF) except control traffic which is directly sent to the IFP. Ingress traffic sent to CPU or transit traffic dropped by policer is still counted whereas egress traffic dropped by QoS is not counted.

Alternatively, it is possible to use the sum of all policer counters for ingress session counters which can be changed in the AAA configuration profile by setting the accounting ingress source to POLICER (default is LIF) to include only accepted transit traffic.

Using LIF in egress and POLICER in ingress results in a symmetric behavior where only accepted transit traffic is counted.

Following the list of the RtBrick class counter attributes.

The internal 64 bit counters are split over the standard 32 bit octet and gigaword counters by using the most significant 32 bit as gigawords and the least significant 32 bits as octets.

The outgoing class counters are filled by queue counters which ensures that only traffic leaving the device is counted here.

Following the list of the RtBrick class counter attributes.

Those attributes will be encoded as subattributes (RFC2865) similar to broadband forum (BBF) line attributes to reduce the size of the RADIUS message. Counters with a zero value will be also excluded from the packet.

The incoming policer attributes count all traffic accepted (colored green) by ingress policers with dedicated packet and byte values per level.

Following the list of the RtBrick policer counter attributes.

Those attributes will be encoded as subattributes (RFC2865) similar to broadband forum (BBF) line attributes to reduce the size of the RADIUS message. Counters with a zero value will be also excluded from the packet.

The underlying virtual output queueing (VOQ) is a common and efficient architecture for high performance switches but brings some major limitations. One obvious limitation is that queues must be allocated in ingress and canΓÇÖt be changed in egress as shown in the picture below.

Service classification and accounting is based on standard behavior aggregate (BA) and multifield classifier (MF) bound global or to ingress interfaces matching on some protocol fields assigning a traffic class. Downstream traffic (to subscriber) is counted per subscriber and class using queue counters to count only traffic not dropped by QoS. Upstream traffic (from subscribers) is countered per subscriber and policer level.

Some service providers would like to control if premium traffic is handled differently on a per subscriber basis which can be also achieved with VOQ as explained below.

Let us assume a product which handles video streaming services with higher priority or excludes those traffic from actual data volume. With VOQ the classification as Video is done on the ingress core interface equally for all subscribers regardless of the booked product. Now instead of changing the queue per subscriber in egress, the behavior of the queue can be changed to behave equally to BestEffort which is supported in egress per subscriber.

The corresponding Video traffic is now threaded equally to BestEffort but still counted separately. For accounting purposes the traffic of BestEffort and Video can be simply added in the billing infrastructure to get the actual amount of BestEffort traffic.

In upstream, the expected class and policer level can be assigned per subscriber because BA and MF classifiers are bound to the PPPoE session in ingress.