Specifications

Page 47 /148
In the threshold option, specify the bucket threshold, which controls the
burstiness of the leaky bucket mechanism. The larger the value, the more bursty
the traffic, which means that over a very short amount of time, the interface can
receive or transmit close to line rate, but the average over a longer time is at the
configured bucket rate. The threshold can be a value from 0 through 65535.
In the overflow option, specify how to handle packets that exceed the threshold:
§ discard--Discard received packets that exceed the threshold. No counting is
done.
§ tag--(receive-bucket only) Tag, count, and process received packets that
exceed the threshold.
If the flow exceeds the bucket's threshold, the packets are either dropped or
tagged, depending on how the receive leaky bucket is configured. If it is
configured to tag packets, the PIC sets the packet-loss priority (PLP) bit in the
notification record associated with the packet to indicate that the packet
encountered congestion in the router. It also indicates that the packet should
have a greater probability of being dropped from the output transmission queue.
2.7 Class Of Service
2.7.1 Implementation of Class Of Service
Route Route
LookupLookup
I/O
Manager
ASIC
PIC
PIC
PIC
PIC
In-bound FPC
I/O
Manager
ASIC
PIC
PIC
PIC
PIC
Out-bound FPC
The Mxxx Series architecture COS mechanisms are implemented on the
incoming PIC port and FPC (I/O Manager ASIC) and on the outgoing FPC (I/O
Manager ASIC). There is also a rate-shaping mechanism on outgoing PIC ports.
There is a token bucket mechanism on E3/DS3 and Packet Over SONET/SDH
interfaces which enables rate policing on input. A threshold is configured for
each physical interface (or logical interface in the case of OC48 or SONET
OC12). If incoming traffic bursts above the threshold limit then the customer can
configure the Mxxx to either drop the packets, or mark them by setting the Packet
Loss Priority (PLP) bit. The PLP bit information is carried with the packet
notification to be used at the outgoing queues in the Mxxx.
The Mxxx supports four queues. Packets can be queued based on the IP
precedence bit settings on the incoming packet. Future releases will support
queuing the logical or physical interface through which the packet was received,
the destination IP address of the packet, or the application type or payload of the
packet.
Packets are pulled from queues and transmitted via a weighted round robin
mechanism. The user can configure the weights for the queues. The amount of
memory allocated to each queue is configurable as well.
Each E3/DS3 or SONET interface (or SONET sub-interface) constitutes a CoS
port. A port supports four queues or classes of traffic. Users can configure the
amount of memory allocated to each of the four queues (I.e. queue length) within
a port, depending on latency requirements (e.g. shorter queue lengths to support
voice traffic). The allocated memory directly affects the RED process.
In addition, users can configure the weights or percentages used by the weighted
round robin mechanism when servicing the four queues. WRR configuration is
port-specific.
The two queue drop profiles and the port drop profile for each FPC is
configurable. The two queue profiles are used to distinguish between packets