Foundry NetIron M2404C and M2404F Metro Access Switches
Configuring HQoS (Rev.03)
QoS/HQoS Implementation
© 2008 Foundry Networks, Inc.
Page 14 of 98
them to the ES Processor (and vice versa). Traffic is sent to the ES Processor via two 1Gb/s
links with load-balancing between them, which makes sure the bandwidth on these links is
fully used and the chances of internal congestion are low. The Packet Processor also performs
bridging of User-to-User traffic and replication of both User-to-User bridged traffic and
multicast traffic coming from the ES Processor. Replication of multi-cast traffic off-loads this
task from the ES Processor and saves bandwidth on the internal connection links.
•
Queuing & WRED/TD
. The Packet Processor supports up to 8 queues per port (a single
queue per FC), with color-aware WRED or Tail-drop for congestion control (both on egress,
towards the User and on ingress, towards the ES Processor). Together with policing, the
queuing/dropping mechanisms help control over-subscription of device internal connection
ports. Color-awareness of the dropping algorithms allows improving traffic handling upon
congestion by first dropping packets belonging to flows (applications/protocols) that are less
sensitive to packet loss.
•
Scheduling & Shaping
. Each port of the Packet Processor (both towards the User side and the
ES Processor) has a hybrid mode scheduler and a single-rate shaper. The scheduler can
schedule the packets out of the queues using SP (with fixed priority per queue), WRR (with
configurable weights) or hybrid algorithm. Hybrid scheduling means that some of the queues
are serviced using SP and the rest are treated as “single priority” by the SP and are serviced
using WRR among them. The shaper is single-rate and allows configuring a rate and a burst
size per queue and per port (2 shaping levels). It is important to emphasize that the scheduling
mechanism used, either strict or WRR, applies also to control packets sent by the local device
and used for the various signaling protocols. Loss of such packets may cause failures in the
network operation.
NOTE
In order to configure the various QoS-related functions of the Packet Processor,
please refer to the following chapters of the User Guide:
•
For Classification, Filtering, Mapping to FC & Color, Policing and
Marking/remarking refer to “Configuring Access Control Lists (ACLs)”.
•
For Queuing, WRED/Tail-drop, Scheduling and Shaping refer to “Configuring
Quality of Service (QoS)”.
The main functions of the ES Processor, as shown in
Figure 8
are:
•
Service Ingress and Network Ingress Mapping to FC + Color
maps the incoming packet to
one of 8 defined Forwarding Classes and the Color (Green or Yellow), which is used by
WRED to prioritize dropping of packets. The mapping of packets coming from the Packet
Processor (Service Ingress) can be done according to DSCP or 802.1p. The mapping of packets
coming from the Enhanced Uplink Ports (Network Ingress) is done according to MPLS EXP.
Please note that the internal FC and Color inside the ES Processor are not necessarily the same
as the ones in the Packet Processor.
•
Service Ingress Policing
(future functionality) will allow an alternative traffic rate-limiting
option using a DRTCM (Dual-Rate Three-Color-Marker) algorithm. This function will serve as
an enhanced alternative to the SRTCM.
•
Remarking
can currently be performed in Network direction only (marking or remarking of
the MPLS EXP field). This marking/remarking is based on both FC and Color, and is always
enabled. When performed on packets sent from the User side to the Network side (Add path), it
can be based on the result of policing, thus notifying the next nodes in the Service Provider’s
network which User packets are non-conforming to the SLA.