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
- 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.
- 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.
- 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.
- In the PoC measurement matrix, what does the delta between
IAM Auth onandIAM Auth offtell 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.
- 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.
- 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.