DenseChip

Interface Protocols · All levels

CHI Topology Basics: Interview Drills

Interview Drills for CHI Topology Basics.

Interview drills

Interview Drills for CHI Topology Basics focuses on request retry rate, directory occupancy, p99 fabric latency. The goal is to connect the observable symptom to protocol mechanism, ownership, and regression risk.

diagram
PROMPT
You see request retry rate, directory occupancy, p99 fabric latency on CHI Topology Basics. Walk through root cause and fix.

STRONG ANSWER
1. Names the layer and transaction identity.
2. Explains CHI separates request, response, data, and snoop flows across nodes with directory and home-node responsibilities.
3. Requests CHI node map, VC utilization report, request/response correlation log.
4. Proposes one reduced sequence and one system regression.

WEAK ANSWER
Jumps to widening the interface, increasing FIFO depth, or blaming firmware without evidence.

Diagram to draw on the whiteboard

CHI node types on a mesh

diagram
CHI MESH (RN = request node, HN = home node, SN = slave node)

   RN-F --- X --- X --- RN-I
    |       |     |       |
    X --- HN-F --- X --- X
    |       |     |       |
   SN  --- X --- HN-F --- SN

RN: cores/accelerators that issue requests
HN: home node owns coherency + directory for an address range
SN: memory/peripheral endpoints

Latency = hops x per-hop cost; placement matters.

Root-cause tree to narrate

diagram
ROOT-CAUSE TREE — CHI Topology Basics

request retry rate, directory occupancy, p99 fabric latency looks wrong
        |
   reproducible?
     /        \
   no          yes
   |            |
 flaky env   same first transaction every time?
 / seed         /            \
              yes             no
               |               |
        protocol rule     timing/reset/PVT
        or config bug     or load-dependent

Protocol deep dive

Coherence extends memory transactions with snoop and state — traffic multiplies when software shares cache lines.

Concept diagram

diagram
COHERENCE TRAFFIC FLOW

RN issues coherent read
   -> HN looks up directory
   -> snoops to sharers
   -> data + state update returned

False sharing: different variables, same cache line -> coherence storm.

Metric graph

diagram
COHERENCY TRAFFIC STACK

data fetch        ████████
snoop responses   ██████████████
writebacks        ██████
maintenance ops   ████

High snoop stack with good IPC -> suspect line sharing before faster NoC.

Metrics and artifacts to collect

  • snoop rate

  • intervention latency

  • coherency transaction mix

  • false sharing indicators

Mini case study

Benchmark IPC looked fine but system power spiked: per-core counters were on one cache line. Padding counters fixed coherency traffic without any NoC change.

Debug branches

  • If snoop latency high, check home node placement and directory policy.

  • If ordering bug, run litmus sequences before microarch changes.

  • If traffic storm, profile cache line sharing in software layout.

Senior review question

Ask: what is the first transaction that deviates, and which spec rule does it test?

Key takeaways

  • Connect every protocol claim to a transaction identity and measurable metric.

  • Store the artifact (waveform, log, counter) next to every signoff decision.

Common pitfalls

  • Debugging timeouts without finding the first bad transaction.

  • Quoting peak bus width without payload efficiency and retry overhead.

  • Treating VIP compliance as a substitute for system integration replay.

Interview whiteboard

Draw layers first, then place the failing transaction on the diagram.

Practice this lesson