QoS Overview

• Provide different priority to different applications.
  • Different service levels for different types or ‘classes’ of traffic flows.

• Cause:
  • Resource contention
    • Multiple flows using same link.
    • Same or multiple applications
    • Each app has its own requirements.
  • Contention results in queuing.
    • Packets may be delayed or dropped.
    • Effective flow throughput decreases.
    • Delay or Jitter may exceed threshold.

• Best Solution
  • Don’t over provision
• Next Best
  • QoS
    • Congestion is controlled.
    • Delay/Loss/Jitter/Throughput are controlled.
    • Only alleviates temporary congestion.

• QoS Models
  • Integrated Services
    • RFC 1633
    • Connection-oriented model.
    • Every flow has an explicit reservation end-to-end.
    • Does not scale well because network must maintain too much state.
    • Best use case is MPLS TE
  • Differentiated Services
    • RFC 2475
    • Connectionless model
    • Traffic is grouped into classes.
    • QoS behavior is defined by traffic’s class.
    • Called Per-Hop-Behavior (PHB)
    • Focus for CCIE

DiffServ:

• Classification and Marking
  • In order for DiffServ to work, traffic must be placed into correct classes.
    • ‘classifications’
  • Traffic classification normally occurs at network ingress edge.
    • Typically a manual process we must enforce.
  • Classification can be encoded inside packet itself.
    • Known as packet’s ‘marking’.

• Classification Types:
  • Classification and marking can happen at multiple places.
  • Layer 2 Class of Service (CoS)
    • 802.1q Ethernet Header
  • Layer 3 IP Type of Service (ToS)
    • IP Precedence and Differentiated Services Code Point (DSCP)
  • Layer 4
    • TCP and UDP Ports
  • Upper Layers
    • Network Based Application Recognition (NBAR)
    • Deep Packet Inspection (DPI)

• QoS Tools:
  • Used to implement QoS Models
    • Many tools rely on correct QoS classification and marking
  • Different Tools for
    • Network Edge
    • Network Core
  • Tools fall into three main categories.
    • Admission Control
    • Congestion Management
    • Congestion Avoidance

• Admission Control
  • Used to enforce traffic marking or traffic rate
  • 2 main types:
    • Traffic Policing
    • Traffic Shaping

• Traffic Policing
  • Used to limit inbound and outbound traffic flows
    • Traffic that exceeds the rate can be dropped, marked, or re-marked.
    • Typically applied on ingress edge.
  • Example use case
    • PE connects to CE with GigE port
    • Circuit is provisioned at 250Mbps
    • PE applies inbound policer at port level
      • If traffic <=250Mbps, transmit
      • If traffic > 250Mbps, drop
• Traffic Shaping
  • Used to normalize outbound traffic flows
    • Smooth out traffic bursts
    • Prepares traffic for ingress policing
    • Delay and Queue exceeding traffic
  • Example use case
    • PE connects to CE with GigE port
    • Circuit is provisioned at 250Mbps
    • CE applies outbound shaper at port level
      • If traffic <= 250Mbps, transmit
      • If traffic >250Mbps, queue for later transmission

• Congestion Management Techniques
  • Used to deal with congestion once it occurs
    • ie. Queueing
  • Queueing Types
    • First in First out (FIFO)
    • Weighted Fair Queueing (WFQ)
    • Priority Queueing (PQ)/Low Latency Queueing (LLQ)
  • Example use case
    • CE to PE link is experiencing packet loss
    • Apply LLQ to give VoIP low delay
    • Apply WFQ to guarantee 50% BW for SQL
    • All other traffic gets best effort FIFO.

• Congestion Avoidance Techniques
  • Stop congestion before it occurs
    • Packet drop strategy
  • Drop strategy types
    • Weighted Random Early Detection (WRED)
    • Tail Drop
  • Example Use Case
    • CE to PE link is experiencing packet loss
    • Apply WRED to selectively drop low priority flows.
    • Senders go into slow start
    • Congestion management is offloaded to end host TCP stack.