Road to CCNP: Day 103

	1.4 Interpret QoS configurations 

Best-effort = no attempt made, once queue is full packets are dropped

IntegratedServices (intserv) = qos built into apps that signal to the endhost and network devices how much bandwidth is needed to function, uses RSVP (resource reservation protocol). This RSVP’d bandwidth cannot be shared, if not used is wasted. All devices in the path must support IntServ. Does not scale well on large networks because of how many reservations would be made and wasted

Differentiated Services (DiffServ) = QoS is managed on a hop-to-hop basis
IP traffic is divided into classes, and classes have different levels of service
Each packet, when received, has its class read and then treated accordingly (like VIP, manager, regular employee, random nobody)

Modular QoS (MQoS) = cisco’s approach to implementing QoS on cisco devices

Class maps = define the traffic classifications, command class-map

Policy maps = provide the actions per class (e.g. for VIPs, prioritize highest), command policy-map
Policy maps at its most fundamentals include: 1. A class, with command class [name] and 2. The QoS action

Service policies = applied to interfaces to determine if its applied on inbound or outbound traffic), command service-policy
Can apply the same policy map to multiple interfaces

1. Identify traffic by creating the class map
2. Create policy action with class map
3. Apply policy map with service policy

"I have a good theoretical/practical understanding of wired QoS, including LLQ, CBWFQ, class-maps/policy-maps, and policing versus shaping."
As someone who took the exam and passed a few days ago, this is all you need to know. Without breaking the NDA, the questions I encountered on QoS were a couple and so very basic, they were out of place in a professional level certification exam. – a reddit post

Policy maps have no effect until applied with command service-policy [input | output] [policy-map-name]

All traffic that doesn’t match a class map is unclassified traffic, and the default for unclassified traffic is best-effort. However, unclassified traffic can have QoS actions applied by default

Classification is predominantly
• Layer 2: Mac addresses and 802.1p/q bits
• Layer 3: DSCP bits in packets, source/destination IP addresses
• Layer 4: TCP or UDP ports
• Layer 7: NGFW inspection/ Network Based Application Recognition (NBAR) identify traffic based on source application

Marking is coloring/dyeing a packet so that it can be distinguished even after other operations
• Internal: QoS groups
• Layer 2: 802.1Q/p class of service bits in the frame
• Layer 3: DSCP bits

802.1Q frame fields: PCP, DEI, VLAN ID

Per-Hop Behavior (PHB): Class selector (CS) so DSCP can be backwards compatible with IP Precedence, Default Forwarding (DF): best-effort (aka normal), Assured Forwarding (AF): guaranteed bandwidth, Expedited Forwarding (EF): Used for low delay
AF > EF > DF

Trust boundary: For best results packets should be colored/dyed AT THE SOURCE or AS CLOSE TO THE SOURCE as possible

Class-based marking configuration: use the set command for a traffic class in a policy-map

Policing: STOP! YOU ARE UNDER ARREST FOR EXCEEDING BANDWIDTH! I AM DROPPING YOU!

Best placed on edge devices to prevent traffic from wasting precious bandwidth on the core

Shaping: Gordon Ramsey: Oh dear how precious, I will buffer and delay your traffic to get your message across

Best placed on the egress of the network – it passed QoS checks anyways, we don’t want to drop it, we want to send it eventually

Markdown – when the traffic police notice bandwidth is exceeded they can do two things:

1. Drop the traffic entirely (arrest)
2. Lower the priority (slow down)

Queuing Algorithms:

1. FIFO = first in first out no differentiation
2. Round robin = Each queue takes a turn sending one packet, no prioritization of traffic
3. Weighted round robin (WRR) = Assign a weight to each queue, they get a proportion of the bandwidth according to the weight (e.g. voice 50%, video 30%, other 20%)
4. Custom queuing (CQ) = Combination of Round robin and FIFO. 16  queues and each queue has FIFO
5. Priority queuing (PQ) = four queues (high, medium, normal, low). Highest queues are ALWAYS served first, and the next queue isn’t served until the higher one is done. FIFO within each queue. Low priority queue can be starved if High is never emptied
6. Weighted fair queuing (WFQ) = The link bandwidth is actually just divided and each queue gets their respective weight % of bandwidth (e.g. audio gets 70% of the link, data gets 30%)