Quality of Service
Course Description
QoS (Quality of Service) is a set of tools that changes how routers and switches treat different types of traffic. By default, a network device forwards packets first come, first served. When a link is congested, that means your voice call competes with a backup job for the same queue. With QoS, we classify and mark traffic, give important traffic its own queue, and limit the bitrate of other traffic with shaping and policing. In this course, we start with the basics like DSCP values and work our way through queuing, shaping, policing, LAN QoS on switches, and RSVP, with Cisco IOS configuration examples.
Course Highlights
In this course you will learn:
- Why we need QoS, and how bandwidth, delay, jitter, and loss affect applications like voice and video.
- What IP precedence and DSCP values are, and how the ToS byte evolved.
- How to classify and mark traffic on Cisco IOS routers with the MQC.
- How CBWFQ and LLQ (Low Latency Queuing) work, and the difference between bandwidth and bandwidth remaining.
- How WRED drops packets before a queue is full to prevent TCP global synchronization.
- How to configure the trust boundary, classification, marking, and queuing on Cisco switches.
- How traffic shaping works with the Tc, Bc, and Be values, and how to configure shape average and shape peak.
- How single rate and dual rate policers work, and how to configure them.
- How RSVP and IntServ reserve bandwidth for a flow.
Presented to you by instructor Rene Molenaar, CCIE #41726
Course at a glance
- 27 lessons in 6 units.
- Format: self-paced lessons with Cisco IOS configurations, iPerf traffic tests, and video.
- Last updated: October 2026.
Requirements
- No QoS knowledge required; we start with the basics.
- Basic knowledge of routing and switching, and how IP packets and Ethernet frames are forwarded.
Course Schedule
What is QoS?
QoS (Quality of Service) is a collection of tools that lets a router or switch give different types of traffic a different treatment. Without QoS, a device forwards everything best effort: one FIFO (First In First Out) queue, and when that queue is full, packets are dropped no matter what they carry. With QoS, we first classify traffic and mark it, for example with a DSCP value in the IP header. Devices further down the path can then use that marking to decide which packets go in which queue, which packets are sent first, and which packets are dropped when there is congestion.
QoS doesn’t create bandwidth. It decides who gets the bandwidth you have when there isn’t enough of it. That matters most for real-time applications. Voice traffic doesn’t need much bandwidth, but it is sensitive to delay, jitter, and loss; a common guideline is a one-way delay below 150 ms, jitter below 30 ms, and less than 1% loss. A data transfer over TCP can recover from a dropped packet, a voice call can’t.
Who uses QoS? Enterprises use it on WAN links and LAN switches to protect voice, video, and business applications. ISPs use policing and shaping to enforce the traffic contract (CIR) with their customers, and customers shape their outgoing traffic so the ISP doesn’t drop it. Most networks use the DiffServ model, where each device makes its own decision based on the marking. The IntServ model with RSVP, where a flow reserves bandwidth end to end, is mostly used for MPLS traffic engineering nowadays.
QoS Learning Roadmap
This roadmap follows the course units. Each topic has a short explanation and links to the lessons where we cover it in detail.
QoS Fundamentals
Before we configure anything, we look at the problem QoS solves. Bandwidth, delay, jitter, and loss each affect applications in a different way, and different applications have different requirements. We also look at the ToS byte in the IP header: how it started with IP precedence and became the DS field with DSCP values. These values are what routers and switches use to recognize marked traffic. The bandwidth delay product tells you how much data has to be in flight to keep a link full.
Classification and Marking
The first step of every QoS policy is classification: identifying the traffic, for example with an access-list or NBAR. Marking then sets a value like DSCP or IP precedence, so other devices don’t have to classify the same traffic again. On Cisco IOS, we configure both with the MQC (Modular QoS CLI): class-maps, policy-maps, and service policies. NBAR adds application layer intelligence, so you can even match on a hostname to block a website. When traffic goes into a tunnel, the original header is hidden, and QoS pre-classify lets you classify on it anyway.
Queuing and Congestion Avoidance
Queuing manages the front of the queue. CBWFQ (Class Based Weighted Fair Queuing) gives each class a minimum bandwidth guarantee with the bandwidth or bandwidth remaining command, and LLQ adds a priority queue that is always served first, which is what we use for voice. The priority queue is policed to prevent queue starvation of the other queues. CBWFQ needs a shaper before you can apply it to a sub-interface. WRED manages the tail of the queue: it drops random packets before the queue is full, which prevents TCP global synchronization.
LAN QoS on Cisco Switches
Even with gigabit links, switches can drop traffic when several interfaces send to one uplink, or when a fast interface sends to a slower one, which is why we need QoS on LAN switches. On switches, we decide where the trust boundary is: do we trust the marking of the IP phone, or do we mark traffic on the access layer switch? Switches do QoS in hardware, so queuing works differently than on routers. We cover the trust boundary, classification and marking, and ingress and egress queuing on Catalyst switches.
Traffic Shaping
An interface can only send bits at its physical clock rate. To get a lower bitrate, a shaper sends traffic for a part of each time interval (Tc) and then pauses, buffering everything that exceeds the rate. The Bc (committed burst) is the number of bits we can send each Tc, and the Be (excess burst) lets us burst after a period of inactivity. You’ll learn how to calculate these values, how to configure shape average and shape peak on Cisco IOS, and how to do frame-relay traffic shaping with the MQC.
Traffic Policing
Policing also limits the bitrate, but instead of buffering, it drops or re-marks the traffic that exceeds the rate. ISPs use it to enforce the CIR (Committed Information Rate) of their customers. A policer sorts packets into conforming, exceeding, and violating, and you decide what happens in each case. We cover single rate two-color, single rate three-color, and dual rate three-color policers, with a configuration example on Cisco IOS.
RSVP and IntServ
DiffServ makes decisions hop by hop, based on the marking. IntServ is different: a flow uses RSVP to request bandwidth from every router on the path, and each router decides with admission control whether it can accept the reservation. This is hard to scale because every router has to keep track of every flow, so today RSVP is mostly used for MPLS traffic engineering. On shared segments, a DSBM provides central admission control.
Other QoS Tools
A few more tools complete the picture. On slow links, a large data packet can delay a small voice packet, and PPP multilink LFI (Link Fragmentation and Interleaving) fixes that by fragmenting large packets. To test your QoS policies, you need traffic, and IP SLA can be used as a simple traffic generator so you don’t need extra tools.
Frequently Asked Questions
QoS only does something when there’s congestion. If a queue never fills up, every packet is sent right away. The catch is that congestion doesn’t only happen when your average utilization is high: short bursts, a fast LAN feeding a slower WAN link, or many interfaces sending to one uplink can fill a queue for a moment, and that’s enough to hurt a voice or video call. If you carry real-time traffic, QoS is cheap insurance. Read our introduction to QoS.
Turn it on if your connection gets congested, for example when someone starts a large upload and your video calls or games start to lag. If your connection is fast enough that it never fills up, QoS has nothing to do. On a home router, the upload direction usually matters most, since that’s the slower link you control. Learn why we need QoS.
It can lower your maximum speed a little. Many home routers shape traffic to slightly below your line rate, so the queue builds up in the router where QoS can manage it instead of at the ISP. Some routers also don’t have enough CPU to inspect traffic at full speed. You trade a bit of peak throughput for lower delay when the link is busy. Learn how traffic shaping works.
Only partly. Your ISP and the rest of the Internet treat your traffic as best effort and usually ignore your DSCP markings. What you can control is your own edge: you can queue and shape traffic leaving your network, so voice goes out first on your own upload link. For end-to-end QoS, you need a private WAN like an MPLS VPN, where the provider honors your markings as part of the contract.
Yes, at different depths. CCNA 200-301 covers QoS concepts like classification, marking, queuing, policing, and shaping, without much configuration. CCNP ENCOR expects you to understand and interpret QoS configurations. Check Cisco’s current exam topics for the exact scope. Start with IP precedence and DSCP.
QoS (Quality of Service) is a set of tools that lets routers and switches treat traffic types differently. We classify and mark traffic, then use queuing, shaping, and policing to decide which traffic gets bandwidth first and which traffic is dropped during congestion. Read our introduction to QoS.
DSCP (Differentiated Services Code Point) is a 6-bit value in the DS field of the IP header, the field that used to be called the ToS byte. Devices use it to recognize marked traffic. For example, voice traffic is usually marked with DSCP EF (46). Learn about IP precedence and DSCP values.
Both limit the bitrate of traffic. A shaper buffers the traffic that exceeds the rate and sends it later, which adds delay. A policer drops or re-marks the traffic that exceeds the rate right away. ISPs often police inbound traffic, and customers shape outbound traffic so it isn’t dropped. Learn how traffic shaping works.