Cisco 3560 About-face Departure Queuing
Cisco 3560 departure queuing improves on the amount concepts of 3550 departure queuing by abacus a
couple of key features. First, 3560 departure queuing creates a apparatus to anticipate chain starvation
of the non-PQ queues—a abeyant affair with the 3550 departure scheduling logic. Second, 3560
queuing adds a abstraction affection that slows bottomward departure traffic, which helps anticipate some types of
DoS attacks and provides the agency to apparatus subrate acceleration for Metro Ethernet implementations.
First, it helps to apperceive a few capacity about the basal departure queuing arrangement on the 3560. Like the
3550, there are four departure queues per interface. Additionally like the 3550, one chain can be configured
as an accelerate queue, although it charge be chain 1 (instead of chain 4 on the 3550). The
classification argumentation is the same, with the departure chain actuality bent alongside by the internal
DSCP, with the centralized DSCP actuality compared to the DSCP-to-CoS map and the consistent CoS
being compared to the Cos-to-queue map—just like it is in the 3550.
This area focuses on the scheduler, bold that frames accept been classified and placed into
the four achievement queues. In particular, the 3560 has two options for the scheduler, both application the
acronym SRR: aggregate round-robin and shaped round-robin. The key differences amid the two
schedulers is that while both advice to anticipate chain starvation back a antecedence chain exists, the
shaped advantage additionally rate-limits (shapes) the queues so that they do not beat the configured
percentage of the link’s bandwidth.
462 Chapter 13: Congestion Management and Avoidance
To see the similarities and differences, it is accessible to anticipate about both options after a PQ and
with two scenarios: first, with all four queues captivation affluence of frames, and second, with alone one
queue captivation frames.
In the aboriginal case, with all four achievement queues captivation several frames, both aggregate and shaped modes
work the same. Both use the agreement of weights for anniversary queue, with the queues serviced
proportionally based on the weights. The afterward two commands configure the weights,
depending on which blazon of scheduling is adapted on the interface:
srr-queue bandwidth allotment weight1 weight2 weight3 weight4
srr-queue bandwidth appearance weight1 weight2 weight3 weight4
For example, with the absence weights of 25 for anniversary chain in aggregate mode, still bold that all
four queues accommodate frames, the about-face would account anniversary chain equally.
The two schedulers’ operations differ, however, back the queues are not all full. Accede a second
scenario, with frames alone in one chain with a weight of 25 (default) in that queue. With shared
scheduling, the about-face would accumulate application this distinct chain with that chain accepting all of the
link’s bandwidth. However, with shaped scheduling, the about-face would agilely delay to service
the queue, not sending any abstracts out the interface so that the chain would accept alone its
configured allotment of articulation bandwidth—25 percent in this scenario.
Next, accede the admittance of chain 1 as the antecedence queue. First, accede a case area queues
2, 3, and 4 all accept frames, chain 1 has no frames, and again some frames access in the departure PQ.
The about-face completes its application of the accepted anatomy but again transitions over to serve the PQ.
However, instead of craving the added queues, while all the queues accept frames cat-and-mouse to avenue the
queues, the scheduler banned the bandwidth acclimated for the PQ to the configured bandwidth. However,
this attached queues the balance rather than auctioning the excess. (In this scenario, the behavior is
the aforementioned in both shaped and aggregate mode.)
Finally, to see the differences amid aggregate and shaped modes, brainstorm that the PQ still has
many frames to send, but queues 2, 3, and 4 are now empty. In aggregate mode, the PQ would send
at abounding band rate. In shaped mode, the about-face would artlessly not account the PQ allotment of the time so
that its all-embracing amount would be the bandwidth configured for that queue.
Hopefully, these examples advice authenticate some of the similarities and differences amid the
SRR scheduler in shaped and aggregate modes. The afterward account summarizes the key points:
■ Both aggregate and shaped approach scheduling attack to account the queues in admeasurement to their
configured bandwidths back added than one chain holds frames.
■ Both aggregate and shaped approach schedulers account the PQ as anon as accessible if at aboriginal the PQ
is abandoned but again frames access in the PQ.
■ Both aggregate and shaped approach schedulers anticipate the PQ from beyond its configured
bandwidth back all the added queues accept frames cat-and-mouse to be sent.
■ The shaped scheduler never allows any queue, PQ or non-PQ, to beat its configured
percentage of articulation bandwidth, alike if that agency that articulation sits idle.
NOTE The 3560 supports the adeptness to configure aggregate approach scheduling on some queues,
and shaped approach on others, on a distinct interface. The alone aberration in operation is that the
queues in shaped approach never beat their configured bandwidth setting.