Schedulers Overview for ACX Series Routers
You use schedulers to define the properties of output queues. These properties include the amount of interface bandwidth assigned to the queue, the size of the memory buffer allocated for storing packets, the priority of the queue, and the random early detection (RED) drop profiles associated with the queue.
You associate the schedulers with forwarding classes by means of scheduler maps. You can then associate each scheduler map with an interface, thereby configuring the hardware queues, packet schedulers, and RED processes that operate according to this mapping.
In ACX Series routers, you can configure more than one strict-priority queue per port.
The hardware services the queues in the descending order of queue numbers marked as
strict priority. All the strict-priority queues are given preferential treatment by the
scheduler as long as their shaping rates (or peak information rates) are not met. Unlike
MX Series routers, the ACX Series routers configured with queues as
strict-high at the [edit class-of-service schedulers
scheduler-name
priority
strict-high] statement hierarchy, the service is based
on queue number and not based on sharing the strict-high queues.
Scheduling on ACX Routers Running Junos OS
Unlike other ACX Series routers, ACX5048 and ACX5096 router supports CIR among
strict-priority queues. There is no implicit queue number-based priority among the
strict-priority queues. Unlike other ACX Series routers, ACX5048 and ACX5096 router
supports configuring drop profiles for loss-priority low,
medium-high, and high for non-TCP protocols as
well.
The options buffer-partition multicast percent
<0-100> at the [edit
class-of-service schedulers scheduler-name
buffer-size] hierarchy level and multicast
<0-100> at the [edit
class-of-service schedulers scheduler-name
shared-buffer-maximum] hierarchy level are supported only on
ACX5048 and ACX5096 routers. For more information, see Shared and Dedicated Buffer Memory Pools on ACX Series Routers.
ACX5448 routers support port-based queueing, scheduling, and shaping. You can
configure up to eight queues (virtual output queues) per physical interface (port).
Scheduling properties can be applied at both physical as well as logical interface
levels. The egress scheduler supports two priority levels
(strict-high and low). Multiple strict-high
priority queues and multiple low (default) priority queues can be configured.
By default a port on an ACX5448 router gets a dedicated buffer of 100 microseconds and shared buffer from DRAM. Delay buffer controls the latency of the queue during congestion and maximum number of packets that can be held in a queue. Default buffer size per port is 100 microseconds.
On all ACX Series routers running Junos OS, the strict priority
queues cannot have transmit-rate configured.
On ACX710 routers, class-of-service commit changes, particularly for schedulers on a physical interface, cause very short (less than one second) pauses in forwarding packets. This can cause packet loss. For some protocols that use very short keepalive intervals, such as BFD, we recommend you disable or enlarge the keepalive interval before you apply scheduler changes. We also recommend you configure scheduling before you enable live traffic.
On ACX 4000 routers, whenever the scheduling and shaping parameters of a port or any of its queues are changed, the entire scheduling configuration on the port is erased and the new configuration is applied. During this window, the traffic pattern does not adhere to user parameters. We recommend you configure scheduling before you enable live traffic.
Scheduling on ACX Routers Running Junos OS Evolved
All ACX7000 Series routers run Junos OS Evolved.
Scheduling Priorities
ACX7000 Series routers support six scheduler priority levels. This support is for both port-level and hierarchical scheduling.The levels are, in order of priority:
-
low-latency
-
strict-high
-
high
-
medium-high
-
medium-low
-
low
ACX routers do not guarantee round-robin distribution between same priority queues.
On all ACX routers running Junos OS Evolved, you can only configure
transmit-rate on low priority
queues.
Low Latency Queuing (LLQ)
ACX7000 Series routers support low latency queuing (LLQ). LLQ enables
delay-sensitive data to have preferential treatment over other traffic. A
low-latency queue has the highest priority over any other
priority queues, including strict-high queues, as well as a low
delay scheduling profile.
For port scheduling of virtual output queues (VOQs), low latency VOQs receive their own dedicated egress queue. High priority VOQs receive a second dedicated egress queue, and low priority VOQs receive a third dedicated egress queue.
Due to the scheduling hierarchy of hierarchical class of service (HCoS), a hierarchical scheduling can use a maximum of two egress queues. Therefore for hierarchical scheduling of VOQs, low latency VOQs and high priority VOQs receive a common dedicated egress queue, and low priority VOQs receive the second dedicated egress queue.
We recommend the following when configuring low-latency VOQs:
-
Use policers to normalize the burstiness of traffic before it reaches a low-latency VOQ.
-
Configure a maximum of two low-latency VOQs on a physical or logical interface.
-
Classify and schedule traffic (that is, reserve bandwidth) for low-latency VOQs so that there is no congestion for those queues.
Low-latency queues receive the same buffers as other queues to efficiently use the limited hardware VOQ buffer profiles.