The scheduling modes interact with the minimum and maximum bandwidth CoS queue and maximum bandwidth egress port shaping specifications. Each egress port may be configured to have a specific scheduling mode. The scheduler first services the queues to meet their CIR and then services the queues to meet the PIR. There are five possibilities as follows:
Strict priority scheduling across CoS queues
The strict priority scheduler provides strict priority access to the egress port across the CoS queue from highest CoS queue index (7) to the lowest (0). The purpose of the strict priority scheduler is to provide lower latency service to the higher CoS classes of traffic. In this mode, the scheduler services the queues in order of their prority in both the CIR and PIR loop.
As described in the following table, CoS queues 7 and 6 each have a minimum bandwidth specification of 10 Mbps, whereas the remaining QoS queues have a minimum bandwidth specification of 50 Mbps. All CoS queues have a maximum bandwidth specification of 1 Gbps.
The goal of these settings is to guarantee the minimum bandwidth settings for each of the queues while also allowing each CoS queue to fully use the egress port capability by having the maximum bandwidth setting at 1 Gbps. The strict priority scheduler mode provides low latency service for CoS queues 6 and 7 while their minimum bandwidth guarantees are being satisfied.
| QoS queue name | Minimum bandwidth | Maximum bandwidth |
|---|---|---|
7 |
10 Mbps |
1 Gbps |
6 |
10 Mbps |
1 Gbps |
5 |
50 Mbps |
1 Gbps |
4 |
50 Mbps |
1 Gbps |
3 |
50 Mbps |
1 Gbps |
2 |
50 Mbps |
1 Gbps |
1 |
50 Mbps |
1 Gbps |
0 |
50 Mbps |
1 Gbps |
Round robin scheduling across CoS queues
The round robin scheduler mode provides round robin arbitration across the CoS queues. The scheduler visits each backlogged CoS queue, servicing a single packet at each queue before moving on to the next one. The purpose of the round robin scheduler is to provide fair access to the egress port bandwidth (at a packet level). This works best when packet sizes are approximately comparable. In this mode, the scheduler services the queues in round-robin for both the CIR and the PIR loop.
Weighted round robin (WRR)
In WRR mode, the scheduler provides access to each CoS queue in round robin order.When the scheduler is providing access to a particular queue, it services a configurable number of back-to-back packets before moving on to the subsequent CoS queue. A value of strict is used to designate that a particular queue be considered to be a part of a hybrid Strict + WRR configuration.
The values 1 to 15 are used to indicate the number of back-to-back packets to be serviced when the scheduler is servicing a particular CoS queue. If the weight specified is N, but if the number of packets in the queue is lesser than N, the scheduler continues working and moves on to the next backlogged queue. In this mode, with no strict queues configured, the scheduler services the queues in round robin in the CIR loop. The configured weights are not considered in the CIR loop. The weights are used only in the PIR loop.
Weighted deficit round robin (WDRR) scheduling
An inherent limitation of the WRR mode is that bandwidth is allocated in terms of packets. WRR works well if the average packet size for each CoS queue flow is known.WDRR aims at addressing this issue. WDRR provides a bandwidth allocation scheduler mode that takes into account the variably-sized packet issue by maintaining sufficient state information when arbitrating across the CoS queues. In this mode, with no strict queues configured, the scheduler services the queues in round-robin in the CIR loop.
The configured weights are not considered in the CIR loop. The weights are used only in the PIR loop. A weight value of 1 to 15 can be configured for each queue. Based on the weights provided respective amount of bytes is de-queued from the queue. A value of 0 is used to designate that a particular queue be considered to be a part of a hybrid Strict + WDRR configuration. If a weight value of 1 is given for queue 1 and 5 is given for queue 2, we see traffic out of the port in the ratio of 1:5 between the queues (1 and 2), provided no traffic is flowing in the other queues. A weight value of 1 will actually pump out 2Kbytes from that queue, a value of 5 will pump out 10 Kbytes. Twice of the weight value given will be pumped out.
Strict + WRR/WDRR
If the WRR/WDRR weight associated with a particular CoS queue is set to strict, the queue is considered to be in a strict priority mode. This set of strict priority queues is serviced first in the order of their CoS numbering (higher numbered CoS queue receives service before smaller numbered queues). In this mode, the scheduler services the strict queues first and then the queues configured with weights in both the CIR and PIR loop. The scheduler ensures that it meets the CIR of all the queues (both strict queues and queues with weight), if bandwidth is available before scheduling the queues in the PIR loop. If multiple queues are configured as strict, the higher-priority strict queues are serviced first before the lower priority strict queues in both the CIR and the PIR loop. The weights configured for the queues are only considered during the PIR loop.