Schedulers on 7210 SAS-Mxp
This section provides information about the scheduler support available in the 7210 SAS-Mxp devices for network port and SAP.
Overview
On the 7210 SAS-Mxp, users have an option to use either port-based egress queuing and shaping or SAP-based egress queuing and shaping for SAPs configured on access ports or hybrid ports. The configure system resource-profile qos port-scheduler-mode command allows you to select the mode to be used for SAPs configured on all the ports of the node (that is, this is a per node setting). The following sections describe the behavior of the scheduler in these two modes.
The queue parameters and scheduling parameters for the queue are configured in the SAP egress policies when using SAP-based egress queues, and are configured in the access-egress policies when using port-based egress queues.
Scheduling with SAP-based queues on access ports
When SAP-based scheduling is enabled, the following figure shows the scheduling for access port, with multiple SAPs configured and when the port-scheduler-mode is disabled. There are 8 egress queues per SAP, a per SAP scheduler and a per access port scheduler.
Each FC/queue for the port can be shaped to configured rates (CIR/PIR). This is used to control the amount of bandwidth allocated to the FC/queue.
FC-to-queue mapping is system-defined and not user configurable.
The queue number determines the priority of the queue, which is used only when the queues are configured as ‟strict.” Queue "8" is the highest priority, and queue "1" is the lowest priority.
On the 7210 SAS-Mxp, only unicast traffic sent out of RVPLS SAPs uses per-SAP egress queues. BUM traffic sent out of RVPLS SAPs uses per-port egress queues. Per-port egress queues are not depicted in the preceding diagram. There are eight per-port queues, and they contend with per-SAP queues for bandwidth.
A queue can be defined to operate in strict mode or weighted mode. The queue mode determines the order of scheduling by the port scheduler.
In a CIR loop, scheduling is packet-based round-robin with a weight of 1.
The behavior of the scheduler for an access port is as follows:
Per port scheduler is available and works at line-rate or configured port egress rate.
Port scheduler distributes the bandwidth available to all the SAPs using WDRR scheduling mechanism (that is, all SAPs have equal weights assigned by the system).
The port scheduler uses the following two passes to distribute bandwidth across SAPs.
CIR loop
The port scheduler distributes the available bandwidth to all the SAP in a round-robin order up to the configured CIR rate (CIR is configured in the aggregate shaper rates for the SAP).
PIR loop
The port scheduler distributes the remaining bandwidth (the bandwidth available after the CIR loop) to all SAPs in a round-robin order (all SAPs are assigned equal weights by the system) up to the configured PIR rate (PIR is configured in the aggregate shaper rates for the SAP).
Each SAP has a per SAP scheduler which operates in SP + WDRR mode and an aggregate per SAP shaper (CIR/PIR). The per SAP scheduler distributes the available bandwidth to the configured strict and weighted SAP queues, using the configured mode and rates, in 2 passes - CIR loop and PIR loop
The CIR loop distributes the available bandwidth (from the bandwidth allocated to it by the port scheduler) to all the queues is in the following order:
higher priority strict queues get the bandwidth up to the configured CIR
any remaining bandwidth, if available, is distributed among the lower priority strict queues up to the configured CIR
any remaining bandwidth, if available, is distributed among the weighted queues (in the CIR loop, weights are not used and therefore the bandwidth is distributed in equal proportion irrespective of weights configured)
The PIR loop distributes the remaining bandwidth (the bandwidth remaining after CIR loop) to all the queues in the following order:
higher priority strict queues get the bandwidth, up to the configured PIR
any remaining bandwidth, if available, is distributed among the lower priority strict queues, up to the configured PIR
any remaining bandwidth, if available, is distributed among the weighted queues in proportion to their configured weights
Each queue can be configured with a queue-mode (strict or weighted) and is associated with a shaper (which allows for configuration of CIR/PIR). The queue mode determines the order of scheduling by the SAP scheduler and shaper rate controls the amount of bandwidth used by the queue.
Scheduling on network ports
For a network port, the scheduling behavior is similar, except that per SAP scheduler is not present in the hierarchy. Instead, per port scheduler distributes the available port bandwidth to all the queues configured on the port in two passes with the behavior being similar to the per SAP scheduler (as mentioned above). Additionally, all the traffic sent out on the network port uses a set of 8 queues which are mapped to the 8 forwarding classes. This is shown in the following figure.
Scheduling on hybrid port with SAP-based egress queues
The following figure shows the scheduling hierarchy for hybrid port with both SAPs and network interfaces configured.
The scheduling behavior is very similar to the SAP-based queues on the access ports. That is, the network port queues used for network traffic are treated as another SAP node in the scheduling hierarchy.
Port-based scheduling and queuing on access ports
Port-based scheduling and queuing shows port-based scheduling and queuing enabled on Access-ports on 7210 SAS-Mxp.
When port-scheduler-mode is enabled, traffic sent out of SAPs configured on access ports and hybrid ports, share a set of 8 egress queues which are mapped to the 8 forwarding classes. A per port scheduler (similar to the one available for network port) distributes the available port bandwidth to all the queues configured on the port in two passes with the behavior being similar to the per SAP scheduler (see below for more details).
Per-SAP scheduler is not present in the scheduler hierarchy when port-scheduler-mode is enabled.
The behavior of the Port-based Queuing and Scheduling on Access Ports is as follows:
If enabled, all SAPs on the node/chassis use port-based queuing.
In other words, user has an option to use either SAP-based queue for all SAPs configured on the node or port-based queues for all SAPs configured on the node. A mix and match of some SAPs using port-based queues and some SAPs using SAP-based queues is not supported.
All SAPs on an access port share the 8 egress queues on the port. On Hybrid ports, SAPs use network port queues. In other words, on hybrid port, all the SAPs configured on the port and the network port IP interfaces share the 8 egress queues on the port.
Supports 2 level hierarchical shaping, with Per queue shaper and per port aggregate shaper (ERL).
Each FC/queue of the port can be shaped to configured rates (CIR/PIR). This is used to control the amount of bandwidth allocated to the FC/queue.
FC to queue mapping is system-defined and not user configurable.
The queue number determines the priority of the queue. Priority of the queue is used only when the queues are configured as strict. Queue ‟8” is the highest priority and Queue ‟1” is the lowest priority.
A queue can be defined to operate in strict mode or weighted mode. The queue mode determines the order of scheduling by the port scheduler.
The scheduling behavior is similar to the one supported on SAPs (modified as below):
Each Access Port has a per port scheduler which operates in SP + WDRR mode and an aggregate per port shaper (ERL). The per port scheduler distributes the available bandwidth to the configured strict and weighted queues, using the configured mode and rates, in 2 passes - CIR loop and PIR loop:
The CIR loop distributes the available bandwidth to all the queues is in the following order:
Higher priority strict queues get the bandwidth up to the configured CIR.
Any remaining bandwidth, if available, is distributed among the lower priority strict queues up to the configured CIR.
Any remaining bandwidth, if available, is distributed among the weighted queues (in the CIR loop, weights are not used and therefore the bandwidth is distributed in equal proportion irrespective of weights configured).
The PIR loop distributes the remaining bandwidth (the bandwidth remaining after CIR loop) to all the queues in the following order:
Higher priority strict queues get the bandwidth, up to the configured PIR.
Any remaining bandwidth, if available, is distributed among the lower priority strict queues, up to the configured PIR.
Any remaining bandwidth, if available, is distributed among the weighted queues in proportion to their configured weights.
Scheduling on hybrid port with port-based SAP queues
The following figure shows the scheduling hierarchy for a hybrid port with SAPs using port-based queues and network IP interfaces using port-based queues. The scheduling behavior is similar to that of access port when port-based queues are used (as described above).
