Analysis of a Quality-Attribute Trade-off Under Cost Pressure
A SaaS provider runs a multi-tenant reporting product. Customers complain about slow report generation at month-end. Analysis shows the bottleneck is a shared query service that all tenants use, with no per-tenant isolation. Engineering proposes either (a) dedicated query instances per enterprise tenant, or (b) a caching layer with query prioritization. Option (a) roughly triples infrastructure cost for the top 20 tenants who generate 60% of load. Option (b) adds cache-invalidation complexity and risks stale financial figures. The CTO wants a decision this week; there is no budget for both. Two enterprise customers have contractual report-latency clauses.
1. Frame the decision in terms of quality attributes and show how the two options trade off.
Read the worked answer
The decision is a quality-attribute trade-off, primarily performance/latency versus cost and consistency (accuracy).
- Option (a) dedicated instances: improves latency predictability and isolation for the top 20 tenants, directly serving the latency clauses. Cost rises roughly 3x for those tenants; it does not improve the long tail.
- Option (b) caching/prioritization: improves average latency at lower cost and helps all tenants, but introduces a consistency/accuracy risk (stale financials) and added complexity (invalidation), which is itself a maintainability concern.
The answer should map each option to the affected attributes and note that the latency clauses are a hard constraint, while cost is a budget constraint-so the analysis must show whether (b) can meet the clauses with acceptable staleness bounds.
2. Recommend a decision path given the one-week deadline and no budget for both.
Read the worked answer
A defensible path:
- Check the clauses first: determine whether they specify a latency bound and whether they tolerate bounded staleness. This single fact may decide the option.
- If clauses are strict on freshness: choose (a) for the contracted tenants only, scoped to the top 20, since that is the smallest spend that satisfies a hard obligation.
- If bounded staleness is acceptable: choose (b) but constrain it-cache only non-financial aggregates, keep financial figures uncached, and set explicit invalidation rules.
- Hybrid within one option: dedicate instances to contracted tenants and apply prioritization to the rest, avoiding a full second program.
The answer should not pretend a week is enough for full analysis; it should identify the one or two facts that gate the decision and commit conditionally.
3. Describe what analysis evidence you would require before finalizing, and how you would present residual risk.
Read the worked answer
Required evidence:
- Load profile: confirm the top 20 tenants cause ~60% of load and model month-end peaks.
- Latency measurements: baseline and projected latency per option against the contractual bounds.
- Staleness analysis for (b): worst-case cache age and which reports would show stale figures.
- Cost model for (a): verify the ~3x figure and whether it scales with tenants or load.
Present residual risk explicitly: for (a), cost overrun and unimproved long-tail experience; for (b), accuracy risk on financial reports and invalidation defects. Recommend monitoring thresholds that would trigger revisiting the decision.
The answer should demonstrate that a fast decision is still an evidence-based one, with named residual risks and review triggers.