What is CoS?

Class of Service (CoS) – also called quality of service (QoS) – is a traffic-prioritization mechanism that classifies ingress packets into forwarding classes and schedules egress traffic so higher-priority traffic, such as voice or business-critical applications, is forwarded ahead of lower-priority traffic when a link becomes congested.

Key takeaways

• Classifies ingress traffic and schedules egress traffic by priority.
• Also called quality of service (QoS) – the terms are interchangeable.
• DSCP and 802.1p bits mark a packet’s forwarding class in the header.
• Rewrite rules can remark CoS values as traffic exits a device.
• Versa VOS applies CoS via QoS profiles, policies, and schedulers.

On this page: Why it matters · How it works · Types · CoS & Versa · FAQ

Why is CoS important?

Not all traffic matters equally once a link runs out of headroom. A video call and a file download sharing bandwidth fail very differently: the transfer just slows, unnoticed, while the call stutters or drops. CoS exists because networks have finite bandwidth, and congestion is a matter of when, not if – on a broadband uplink, an aggregation link, or a WAN edge.

Without CoS, a device drops packets essentially at random once a queue fills, and every application degrades together. With CoS, ingress traffic is sorted into forwarding classes on arrival, and egress scheduling ensures voice, video, and business-critical classes are served first, while bulk transfers absorb the delay. That is why CoS underpins SD-WAN, VoIP, and any network carrying real-time and bulk traffic on shared links.

What is CoS?

CoS works in two stages. On ingress, a device examines each packet and classifies it – by a Layer 3 field (DSCP or legacy IP Precedence), a Layer 2 field (802.1p), the source application, or an access-control match – and assigns a forwarding class and loss priority. The class sets which outbound queue the packet uses; the loss priority sets how aggressively it drops as that queue nears capacity.

On egress, the device schedules transmission across queues so higher-priority classes go first, using disciplines such as strict priority or weighted round-robin, plus drop mechanisms such as weighted random early detection (WRED) or tail drops to manage congestion before it becomes loss. Devices can also rewrite the DSCP or 802.1p bits on the way out, carrying the marking into the next hop – important where every device must agree on priority.

Types of CoS

CoS markings live in a few standard header fields, and most networks combine more than one. IP Precedence is the original 3-bit field in the IPv4 Type of Service byte – coarse, with only eight values, and largely superseded. DSCP replaced it with a 6-bit field from the IETF’s Differentiated Services architecture, giving up to 64 markings and classes such as expedited and assured forwarding. IEEE 802.1p marks priority at Layer 2, in the 3-bit field of an 802.1Q VLAN tag, for devices that switch rather than route. On MPLS networks, the EXP bits carry CoS across label-switched paths, where the IP header is not always inspected.

Because these fields sit at different layers, a packet’s priority is often mapped between them crossing Layer 2, Layer 3, and MPLS boundaries – why classifier and rewrite rules exist at each hop.

CoS & Versa

Versa Operating System (VOS) implements CoS end to end. On ingress, QoS profiles define policers and assign traffic to a forwarding class and loss priority, applied through QoS policies (Layer 3/4 match) or application QoS policies (Layer 7); the two can stack, with the application policy’s class taking precedence. On egress, rewrite rules remark DSCP or 802.1p bits, drop profiles configure WRED or tail-drop behavior, and schedulers in scheduler maps control how classes share bandwidth.

For SD-LAN, classifier profiles apply the same DSCP- or 802.1p-based classification to NPU interfaces and networks, with default mappings out of the box. High-throughput platforms – CSG5000 series appliances with Intel E810 NICs, from Release 22.1.1 – support hardware-based shaping at up to 100 Gbps, trading some scheduler flexibility for line-rate performance.

Frequently asked questions

What is the difference between CoS and QoS?

The terms are used interchangeably in most networking documentation, including Versa’s. Some vendors use CoS narrowly for the marking step and QoS for the broader discipline including shaping and policing, but both describe the same concept.

How does a device classify traffic into a forwarding class?

On ingress, the device matches each packet against configured criteria – a DSCP or 802.1p value, address, zone, or application – and assigns a forwarding class and loss priority, which determine the egress queue and scheduling treatment the packet gets.

What is the difference between DSCP and 802.1p marking?

DSCP is a 6-bit field in the IPv4 or IPv6 header, giving up to 64 priority values at Layer 3, so it survives routing between subnets. 802.1p is a 3-bit field in the 802.1Q VLAN tag at Layer 2, so it only applies within a switched, VLAN-tagged segment.

What happens when a rewrite rule changes a packet’s CoS marking?

A rewrite rule reads the packet’s forwarding class and loss priority and writes a new DSCP or 802.1p value into the outgoing header before the packet leaves. This lets a device remark traffic to match the next network’s policy, or align inner and outer tunnel headers.

Does CoS guarantee bandwidth to every traffic class?

No – CoS prioritizes traffic during congestion; it does not reserve bandwidth on links that are never full. Schedulers can guarantee a minimum rate per class, but with no contention, traffic simply passes through regardless of class.

More on CoS

If you want to learn more, these are sources we trust.

Related terms

If you’re reading about CoS, these are worth looking at next.