High Scale Ethernet MDA Capabilities
Quality of Service Guide 767
• Ensure that the policing queue is not stalled by the queue groups configured
aggregate rate limit. If the queue is not the highest scheduling priority, the sum of the
allowed scheduling for the queues with higher scheduling priority may cause the
queue groups PIR to be exceeded and thus scheduling for the policing queue will
stall. If the queue is at a high enough priority, lower priority queues will only be
allowed to consume the group PIR while the higher priority queues are inactive
(empty). In the event that lower priority queues cause the PIR to suspend scheduling
for the queue group, higher priority queues will have first access to ingress
scheduling once the group PIR decrements below the threshold.
• Care should also be taken with ingress secondary shapers. If the queue is assigned to
an ingress secondary shaper, the queue may be stalled when the aggregate rate of the
queues associated with the shaper exceeds the shapers PIR.
Packets sent to an HSMDA ingress queue configured for policing cannot have the ignore-CIR
flag set. Color-aware profiling and ingress queue policing should not be mixed. In the event
that a packet is classified as in-profile or out-of-profile while ingress queue policing is
configured, the ignore-CIR bit will automatically be reset to zero (the queue CIR will be
updated by the packet).
Two implementation options exist to account for out-of-profile scheduling discards:
1. When a packet is discarded due to out-of-profile scheduling, the out-of-profile packet
and octet forward counters within the counter ID associated with the packet are
updated. If the packet is forwarded, the in-profile packet and octet forward counters
within counter ID’ are incremented. Software must add the out-of-profile forwarded
counter to the low priority discard counter to determine the total discards based on
out-of-profile for the queue.
2. Alternatively, the discard event may be hard-coded to increment the discard counters
directly. When discarding, the congestion-priority bit is used to determine whether
the high or low discard counter is incremented.
For packets discarded at enqueuing time, the high priority or low priority packet and octet
discard counters are updated. The decision to use the high or low priority discard counter is
driven by the congestion-priority’ bit associated with the packet. This bit is set based on both
color aware profiling and ingress priority of the packet. If the packet is explicitly classified
as in-profile, the bit is set to high. If the packet is explicitly classified as out-of-profile, the bit
is set to low. If the profile of the packet is undetermined (not explicitly in-profile or out-of-
profile), the bit is set to high or low based on the classified ingress priority of the packet.
To Next Page
Loading...
