Skip to content
DenseChip
Learn
PracticeCode labInterview

DenseChip

Navigate

Menu

LearnPracticeCode labInterview

Tools

All toolsSV editorCode repoResume
DenseChip

Practice, courses, and chip-design tools in one focused workspace.

LearnPracticeCode labInterviewToolsLearning paths

© 2026 DenseChip

AboutContactPrivacyTerms
Skip to lesson

Course contents

  • How to Use This Course
  • Reading Path
  • Cache Coherency Glossary
  • The Shared Course Model
  • Cache Coherency Revision Sheet
  • Why Coherence Exists8
  • States and Protocols12
  • Transactions12
  • Races and Controller Design8
  • Progress and the Fabric16
    • Virtual Networks and Progress
    • Filtering and Snoop Pruning
    • Latency, Fairness and QoS
      • Anatomy of a Miss Latency
      • Fairness on a Contended Line
      • QoS Classes and Coherence
      • Starvation Cases
    • Hierarchical and Chiplet Coherence
  • Ordering and Synchronization12
  • Performance and Scalability12
  • Real Protocols12
  • SoC Integration12
  • Power, Clocking and Reliability12
  • Verification12
  • Debug and Bring-up12
  • Interview and Design Review17
  • Capstone: A Coherent Model Lab5
Loading lesson

‹Progress and the Fabric
  1. Topics
  2. Cache Coherency
  3. Progress and the Fabric
  4. Latency, Fairness and QoS

Part 5 · Progress and the Fabric · All levels

Latency, Fairness and QoS

Separate a transaction's critical path from arbitration policy and conditional progress guarantees.

  • 4topics
  • 4lessons

A fast average does not prove fair service, and fairness does not provide a cycle bound by itself. First account for the serial path of one upgrade, then ask which requests stay eligible and what makes their completion bounded.

Lesson

After it you can...

Anatomy of a Miss Latency

Build an illustrative event timeline without double-counting overlap.

Fairness on a Contended Line

Compare service policies under explicit eligibility assumptions.

QoS Classes and Coherence

Keep priority from violating permission or completion obligations.

Starvation Cases

Find the stage where a live request is bypassed.

Read after resource deadlock and acknowledgement tracking. The examples retain H-collected acknowledgements. Every timing number here is illustrative, not a measurement or a vendor guarantee.

Check yourself Can a fair arbiter guarantee a 20-cycle transaction if response delivery has no upper bound?

No. Fairness can ensure eventual selection under its conditions. A numeric completion bound also needs bounds on service, delivery, delay caused by competing traffic, and endpoint acceptance (the receiving node taking the message into a buffer).

Related topics

  • Shared course model

  • Prerequisite: progress assumptions

PreviousNext: Anatomy of a Miss Latency

Up next

Up nextAnatomy of a Miss Latency

Sign in to save progress

Practice this lessonReport an error

See also

  • Virtual Networks and Progress
  • Filtering and Snoop Pruning
  • Hierarchical and Chiplet Coherence