Weighted Fair Queuing (WFQ)

#fair-queue <CDT> <N Flow Queues> <N Reservable Queues>

• The number of reservable conversations (queues) is the number of flow-queues available to RSVP reservations (if any). Those flows have a very small weight value, and thus are preferred above any other flows. 
• In addition to reserved flow queues, there are special “Link Queues”. 
• The number of queues is fixed to 8, and they are numbered right after the maximum dynamic queue (e.g. if there are 32 dynamic queues, “Link Queues” start at number 33). 
• WFQ uses those queues to service routing protocol traffic and Layer 2 keepalives – everything that is critical to router operations and management. 
• Each queue has a weight 1024, which is lower than any dynamic queue can get, so control plane traffic has priority over regular data traffic.
• WFQ uses an intelligent congestion management solution that provides “fair”sharing of the interface bandwidth between multiple traffic flows.
•  A traffic “flow”(or conversation) is a unidirectional sequence of packets, 
• defined based on the 
• protocol type, 
• the source/destination IP addresses,
• the source/destination ports numbers (when available), 
• and partially on the IPv4 ToS byte value.
• The term “fair” on WFQ refers to the max-min fairness.
• IOS implementation of WFQ assigns weights automatically based on the IP Precedence (IPP) value in the packet’s IP header. 
• The formula is Weight=32384/(IPP+1), where IPP is the IP Precedence of the flow.
• Specifically WFQ implements its fair sharing logic as follows. 
• First, the scheduler creates a group of flow queues for the interface, based on a manual setting or an automatic calculation derived from the interface bandwidth. 
• When a new packet arrives for output scheduling, the scheduler applies a special hash function to the packet's source/destination IP/Port values to yield the flow queue number. 
• This is why the number of queues is always a power of 2, because the hash the output value is somewhere between 0 and 2N.
• This procedure also means that multiple flows may share the same queue, called "hash collision", when the number of flows is large.
• This “virtual time” is  the total size of all packets stored in the flow queue scaled by the flow computational weight.
• Each flow queue has a special virtual scheduling time assigned  the amount of time that would take to serialize the packets in the queue across the output interface.
• Note that the weight is inversely proportional to IP precedence, and thus more important packets have smaller virtual scheduling time value (WFQ thinks that it can serialize them “faster”). 
• It is more appropriate to call this computational weight the “scaling factor” to avoid confusion with the classic meaning of weight.
• 
  
• 
  
•