ONT Chapter 4 – Conegstion Management and Queuing
Also available as a PDF here
Introduction to Congestion Management and Queuing
- Congestion occurs when the rate of input to an interface exceeds the rate of output
- Speed mismatch problem – when traffic on a high speed interface enters the router but exits on a low speed interface
- Aggregation problem – when traffic from multiple interfaces aggregates into a single interface
- Tail drop – occurs when buffers fill up and new incoming traffic is dropped
- Default queuing method is FIFO (first in first out)
- Queuing mechanisms consist of hardware and software components
- Hardware queue is called the transmit queue (TxQ)
- If TxQ becomes full, packets are held in the software queue and release to the TxQ based on the queuing mechanism
- TxQ is always FIFO
- Software queuing has a number of queues – one for each class of traffic
- If the software queue is full, the packet is dropped
- If TxQ is full, packet is held in software queue based on it’s class
- Even if a packet is in one of the software queues, it can still be dropped if WRED is applied to that queue
- The IOS determines the hardware queue based on the configured bandwidth on the interface
- You can manually set the queue size with the tx-ring-limit command
FIFO, Priority Queuing, Round Robin, Weighted Rounded Robin Queuing
First in First out (FIFO)
- Default queuing mechanism on all interfaces > 2Mb
Priority Queuing
- Priority queue has 4 queues (predefined and pre-created)
- High (20 packet capacity)
- Medium (40 packet capacity)
- Normal (60 packet capacity)
- Low (80 packet capacity)
- If packets are unassigned, they go to the normal queue
- If the high priority queue has packets, only the high priority queue has packets being forwarded, none of the other queues are moving packets
- If high priority queue is empty, it checks the medium queue. If medium queue is empty, it checks the normal queue, etc.
- After 1 packet exits any queue, the high priority queue is checked again
- Potential to starve lower priority queues
- Often recommend for lower-bandwidth interfaces where you want to give absolute priority to certain traffic
Configuring a Priority Queue
Create the priority list and set the default to low
router(config)#priority-list 1 default low
Show the queuing priority (only shows what was changed from default)
router#show queuing priority
Configure all traffic coming in on an interface to a specified queue
router(config)#priority-list 1 interface s0/0 low
Assign a priority list to an interface
router(config-if)#priority-group 1 (or whatever number priority-list was created)
Round Robin Queuing
- RR schedules 1 packet from each queue and then a packet from the next, etc.
- No queue has priority over others queues
- If a queue has larger packets then the other queues, that queue may take up more bandwidth
- Effectively there is no mechanism for prioritization
Weighted Round Robin
- Similar to RR, but you can assign a weight to a queue
- Each queue receives a portion of the interface bandwidth
- Not necessarily equal to the others
- Common queuing (CQ) – example of WRR where you can configure the number of bytes to be processed before moving to the next queue
- If byte count (weight) is too close to the MTU of the interface, that queue might take up more bandwidth than intended
- If the weight is much greater than the MTU, queuing delays will increase
WRR Commands
Create a custom queue list (1) and create a queue (1) and match an ACL to put traffic into that queue – (a queue list contains queues – For example, we create custom queue list 1 which contains 4 queues)
router(config)#queue-list 1 protocol ip 1 list ACLNUMBER
Apply the custom queue list created above to an interface
router(config-if)#custom-queue-list 1
Show custom queue list info
router(config)#show queuing custom
Weighted Fair Queuing (WFQ)
- Default queuing on serial interfaces < 2.048 Mbps
- Basis for CBWFQ and LLQ
- Benefits include
- Divides traffic into flows
- Provide fair bandwidth allocation to active flows
- Provide faster scheduling to low-volume interactive flows
- Provide more bandwidth to higher priority flows
- Packets dropped from “aggressive flows” first
WFQ Classification and Scheduling
- Arriving packets are classified into flows – flows are identified by:
- Source IP
- Destination IP
- Protocol number
- ToS
- Source port
- Destination port
- WFQ generates as hash based on these fields – WFQ places packets with the same hash value (in the same flow) into a queue based on the hash value
- If no queue exists, a new one is created
- If all flows have the same weight, WFQ divides the bandwidth equally
- Packets in a high volume queue may end up being dropped
- WFQ dynamically creates and destroys queues – 256 max by default
- Does not offer precise control of bandwidth allocation between flows
- Does not work with tunneled or encrypted packets – WFQ needs access to original IP header info to form the hash
- If number of flows > number of queues available, new flows are assigned to existing queues – multiple flows might end up sharing the same queue
WFQ Insertion and Drop Policy
- WFQ has a queue for all packets of all flows – this is the sum of all the memory taken by packets present in the WFQ system
- Aggressive dropping — If hold queue is full, and a packet arrives – the packet is dropped
- Unless the packet is assigned to an empty queue
- Each flow has a congestive discharge threshold (CDT)
- WFQ early drop — If a packet arrives and the hold queue is not full but the CDT of that flow is reached, the packet is dropped
Benefits and Drawbacks of WFQ
- Benefits include
- Easy to configure
- Does not starve flows
- Guarantees throughput to all flows
- Drops packets from most aggressive flows and provides faster service to nonaggressive flows
- Standard and simple queuing mechanism – support on most Cisco platforms
- Limitations and Drawbacks
- Classification and scheduling are not configurable
- Supported only on slow links (< 2.048 Mbps)
- Cannot guarantee bandwidth or delay
- Multiple flows may be assigned to same queue
- Cannot be used on serial interfaces using:
- Virtual interfaces, including loopbacks and dialer interfaces
- Bridging or tunneling
- LAPB, X.25, SDLC
WFQ Commands
Enable WFQ on an interface
router(config-if)#fair-queue
Change number of dynamic queues (to 64 – must be in power of 2)
router(config-if)#fair-queue 100 64 200 (100 is the congestive discard threshold and 200 is the number of reservable queues)
Show queuing info
router#show queuing interface s0/0
Class-Based Weighted Fair Queuing
- Allows creation of user defined classes – each assigned its own queue (up to 64 queues)
- Each queue receives its own minimum bandwidth guarantee, but can use more if it becomes available
- Each queue is FIFO with a defined bandwidth guarantee and maximum packet limit
- If queue is full, tail drop may occur
- Can be offset with WRED
- Default queue is “class-default” – packets that don’t match any other queues are put here
- Can configure class-default queue to be WFQ, but not the other queues (only in 7500 series routers)
- Class-default queue uses remaining bandwidth (unless specified)
- Does not address low-delay requirements
Classification, Scheduling, and Bandwidth Guarantee
- Accomplished using the MQC using class maps
- Scheduling and bandwidth guarantee offered to each queue is based on the weight assigned to it
- Weight is computed by IOS based on the values you enter for:
- Bandwidth, bandwidth percent, or bandwidth remaining percent
- Bandwidth
- Allocates (reserves) a certain amount of bandwidth (in Kbps) to the queue of a class
- Default maximum reserved bandwidth on an interface is 75% of the total bandwidth of that interface – can be modified using max-reserved-bandwidth command
- Bandwidth percent
- Allocates an amount of bandwidth equal to a certain percentage of the interface bandwidth
- Bandwidth remaining percent
- Allocates a percentage of the remaining available bandwidth to the queue
- Note: You cannot use the bandwidth on one queue and bandwidth percent on another queue – it’s either one or the other on all queues
- Available bandwidth = (Interface bw x max reserved bandwidth) – (sum of all existing reservations)
- Weight is computed by IOS based on the values you enter for:
Benefits and Drawbacks of CBWFQ
- Benefits include:
- Allows creation of user defined classes
- Allows allocation of bandwidth for each traffic class based on user policies/preferences
- Finer granularity than WFQ
- Drawbacks include:
- No suitable queue for real-time traffic – real-time applications expect low-delay queue in addition to bandwidth guarantee
Configuring and Monitoring CBWFQ
- Define traffic classes using class maps
- Create a policy map and link class maps to it
- Define the bandwidth, queue-limit (packet limit), etc. for each queue (class)
Low-Latency Queuing
- Includes a strict-priority queue that is given priority over other queues
- Strict-priority queue is policed, so that other queues won’t starve – meaning it cannot transmit more data than its bandwidth permits
- If more traffic arrives than the strict-priority queue can transmit, the traffic is dropped
- LLQ is essentially CBWFQ with one or more strict-priority queues added to it
- If more than 1 strict-priority queue exists, the queues are policed separately, but after policing, the traffic is not separated
Benefits of LLQ
- Offers all benefits of CBWFQ
- Offers ability to define a high priority queue (strict-priority)
- Strict-priority queue is policed (Unlike PQ)
- Strict-priority queue can accept any traffic, not just RTP
