Skip to content

Latency Impact Analysis Quiz

This quiz tests your understanding of the degrading and improving latency factors in a Lattice migration and how to design a PoC measurement.

Multiple Choice Questions

  1. Why can't the latency impact of a Lattice migration be stated in advance as "it adds N milliseconds"?
    • A) AWS does not publish latency figures
    • B) Proxy work, network paths, TLS reuse and authentication change together, so measure the actual workload
    • C) Lattice is not yet GA
    • D) Latency is determined solely by region
Show Answer

Answer: B) Proxy work, network paths, TLS reuse and authentication change together, so measure the actual workload

Explanation: No Lattice latency measurements are supplied here. The direction and magnitude depend on the tested configuration; do not present microsecond estimates or guaranteed improvement as observations.

  1. What does it mean that p50 and p99 have different factor compositions?
    • A) p99 is always worse than p50, so looking at p50 alone is enough
    • B) The median and tail can react differently to contention, cold starts and routing, so measure both
    • C) p50 and p99 are affected by the same factors, so measuring one is enough
    • D) p99 is noise and should be ignored
Show Answer

Answer: B) The median and tail can react differently to contention, cold starts and routing, so measure both

Explanation: Different percentile responses are hypotheses to investigate, not a guarantee that p50 worsens while p99 improves. Keep errors, throughput and workload conditions alongside the distributions.

  1. Which client setting was noted as potentially mattering more than one proxy hop?
    • A) DNS cache TTL
    • B) keepalive and connection pool settings
    • C) Log level
    • D) Request timeout value
Show Answer

Answer: B) keepalive and connection pool settings

Explanation: The cited benchmark shows that connection reuse mattered in that workload. New TCP connections add setup; TLS handshakes apply when TLS is used. That is not a Lattice latency measurement.

  1. In the PoC measurement matrix, what does the delta between IAM Auth on and IAM Auth off tell you?
    • A) The cost of cross-AZ traversal
    • B) The combined effect of signing, credential handling and enabled-policy evaluation under matched conditions
    • C) The effect of removing Envoy CPU contention
    • D) The pure effect of the path change
Show Answer

Answer: B) The combined effect of signing, credential handling and enabled-policy evaluation under matched conditions

Explanation: Auth on/off does not isolate HMAC alone. Repeat matched runs, keep load and connection settings constant, and report credential refresh, failures and throughput.

  1. Which statement about cross-AZ traffic is correct?
    • A) It is both a latency degradation factor and a separate additional billing factor
    • B) It is a latency degradation factor, but traffic through Lattice incurs no separate inter-AZ charge — it is included in the data processing charge
    • C) It affects neither latency nor billing
    • D) Lattice always selects a Target in the caller's AZ, so it need not be considered
Show Answer

Answer: B) It is a latency degradation factor, but traffic through Lattice incurs no separate inter-AZ charge — it is included in the data processing charge

Explanation: Cross-AZ increases latency due to physical distance (measured baseline: same-AZ 0.339 ms vs cross-AZ 0.544 ms), but on the billing side Lattice does not charge a separate inter-AZ fee — it is folded into data processing. D is wrong: whether Lattice's Target selection considers the caller's AZ is not confirmed in official documentation and must be measured in a PoC, which is exactly why the matrix includes an AZ axis.

  1. Why is warm-up needed in the measurement design, and why should first-request latency still be recorded separately?
    • A) Warm-up is unnecessary; measuring the first request is enough
    • B) Cold credentials or connection pools can add setup work, so record first-request and warmed-up latency separately and evaluate both for frequent cold starts
    • C) Warm-up improves only p50 and has no effect on p99
    • D) The first request always fails
Show Answer

Answer: B) Cold credentials or connection pools can add setup work, so record first-request and warmed-up latency separately and evaluate both for frequent cold starts

Explanation: A first request can include credential acquisition and connection setup when caches or pools are cold. The acquisition path depends on the provider: IRSA uses STS, while EKS Pod Identity uses the node agent and EKS Auth. TLS handshake work applies only to TLS connections. Record first-request and warmed-up latency separately with provider/cache/transport state; do not assume every first request makes a direct STS call or establishes TLS. Evaluate both conditions when cold starts are frequent.