Your short link goes viral

Find the bottleneck during a traffic spike, test a warm cache, then challenge the same fix with cold traffic.

Lesson 1.0.0 · Model 0.2.0

THE BRIEF

A decision with consequences.

A public short link receives 120 redirect requests per second. A popular post suddenly raises demand to 600 requests per second.

Predict the first constrained resource, compare the spike with a cache change, and explain why the change depends on reuse.

Keep database capacity at 200 reads per second for the first comparison.
Use the same seed and duration when comparing runs.
Describe a remaining limitation before calling the change successful.

Objectives and assumptions

Distinguish offered request rate from completed throughput and resource capacity.
Explain why scaling the application cannot remove an unchanged database bottleneck.
Use cache hit and miss evidence to defend when read offload helps.
These are synthetic teaching inputs, not measured Redis, PostgreSQL or cloud performance.
One database service slot handles 200 reads per second. Its FIFO waiting buffer holds 100 requests; excess arrivals are rejected.
The application and network are nonbinding for these presets; additional app replicas do not increase database service capacity.
Cache hits follow a seeded authored pattern. A hit-rate setting does not model keys, TTL, eviction, invalidation or a real warming process.
The hit-rate parameter is a probability; exact finite-run hit counts and database demand come from the seeded trace.
A cold preset represents zero reuse throughout the run, rather than an implemented cache warm-up algorithm.
Runs use a fixed arrival window and report remaining in-flight work explicitly. Successful latency does not include rejected requests.
Keyed presets additionally implement the declared keyspace, LRU/TTL and initial cache state. Those mechanisms apply only when keyedCache is present.

01 · PREDICT

Before you run.

At 600 requests per second without cache, which resource becomes the sustained bottleneck in this declared model?The database: demand exceeds its 200 reads/s service capacityThe application: another app replica should solve itThe network: the link becomes saturatedI'm unsure — I want to inspect the evidenceWhy do you expect that?

Unsure is a useful starting point.

02 · EXPERIMENT

Change one assumption.

Ready

Experiment parameters

Optional mechanisms

Configure arrival distributionModel keys, TTL and evictionAdd a token bucketChoose load-balancing policy

Keyed caching takes effect when the read cache is enabled. Each mechanism changes the declared model and will appear in comparisons.

Application replicasRequests / secondEnable read cacheAuthored cache-hit probability (0–1)DB service / secondDB waiting placesArrival window (ms)Request deadline (ms)Workload seed

Record a prediction first; Unsure is valid.

03 · EXPLAIN

Defend the trade-off.

State the cause, cite recorded evidence, explain the change, and name its downside.

Name the resource and its modeled capacity.
Use a queue, rejection or throughput observation from the run.
Explain what changed and why it changed database demand.
State the cache reuse assumption and one case where the fix fails.

Your reasoning

04 · TRANSFER

Change the requirement.

A product catalog receives the same 600 reads/s, but nearly every request asks for a different uncached item. The database stays at 200 reads/s. Would your previous cache setting still protect it? Propose one defensible response and state what that response gives up.

Requests now have little reuse; do not assume 80% hits.
The database service capacity remains unchanged.

Adapt your answer

Reasoning review rubric

Causal diagnosis: Identify the constrained resource and connect it to offered demand, queue behavior or rejection evidence.
Use evidence: Refer to actual request accounting, cache misses, queue observations or trace events from the selected run.
Defend a trade-off: Explain a downside, unmet requirement or workload assumption associated with the proposed change.
Adapt to the new case: Respond to the changed constraint instead of repeating the previous architecture without justification.

Your reflection on the rubric (optional)

This records your own judgment. It is not an independent assessment.

Causal diagnosisUse evidenceDefend a trade-offAdapt to the new case

What would make this answer fail?

Enable the cache but select the cold preset with zero authored hits.

Expected observation

Database demand remains approximately the offered redirect demand. A cache icon alone does not remove the bottleneck.

Sources, publication, and review

Original project-authored scenario, wording, parameters and feedback. No book text or repository diagrams are bundled.

Content status: published. Recorded technical reviews: 1. Model simplifications are part of each experiment’s contract.

Scaling topic notes — Topic reference only; no commercial reuse permission is inferred from public availability.
SimPy shared resources — Reference for limited-capacity resources and waiting work; the application uses its own teaching model.

Review your practice Try an interview prompt