Queuing works only for packets in the outbound (egress) direction. The only viable way to improve this situation is to enable bidirectional queuing on adjacent routers—for example, by configuring committed access rate (CAR)/rate limits on Cisco IOS architecture. Adjacency essentially means connected via point-to-point links or Ethernet crossover links. When no other queuing regimes are activated, almost all stack implementations resort to default first-in/first-out (FIFO) behavior.
The actual tasks involved in traffic shaping and implementing QoS include reserving resources, buffering, and scheduling behind the scenes. The choice of a queuing discipline is tricky, depending on the load on a link, and requires a thorough understanding of the internal workings of the queuing mechanism.
Queuing disciplines are a classical area of academic and applied research and go beyond the scope of this book. They are essentially procedures or measures that influence the way data is sent, delayed, and queued. You will find excellent resources for further information about queuing in the "Recommended Reading" section at the end of this chapter.
Queuing disciplines essentially come in two flavors: classless queuing disciplines (no subdivision granularity; reschedule, delay, or drop on a flat scale) and classful (class-context) queuing disciplines. The most popular queuing regimes are as follows:
-
CBQ— Class-based queuing
-
RED— Random early detection
-
WFQ— Weighted fair queuing
-
PRIQ— Priority queuing
Permanently saturated links require other strategies than bursty traffic patterns. Nothing really prevents permanently overburdened queues and interface buffers from dropping datagrams/frames. Proactively dealing with that problem is the art of congestion avoidance/management. Queuing is an integral part of the IP stack and forwarding engine and, therefore, the responsibility of the kernel. User-space utilities for administration complement the implementations. Shaping serves two purposes: limiting available bandwidth, and smoothing the use of virtual pipes.
Traffic conditioning is the art of dealing with incoming (ingress) traffic via a policer or shaper. A policer just enforces a rate limit, whereas a shaper smoothes the traffic flow to a specified rate by the use of buffers. Standard mechanisms of the Cisco IOS architecture are CAR, generic traffic shaping (GTS), and Frame Relay traffic shaping (FRTS).
0 comments:
Post a Comment