Design Cost-Optimized Architectures practice questions

Archive-class selection cross-checks two numbers: the retention period against each class's minimum storage duration, and the stated retrieval tolerance against each class's restore speed.

Retention is 120 days and investigators tolerate a wait of several hours. Glacier Flexible Retrieval is the cheapest class that clears both checks: its 90-day minimum fits inside 120 days, and its asynchronous restores in minutes to hours match the tolerance. Deep Archive looks cheaper per GB but its 180-day minimum exceeds the 120-day retention, so every object is billed for 60 phantom days, which is exactly the trap the question tests. Standard-IA and Standard both work functionally but over-deliver millisecond access at higher storage prices, and with rare access there are almost no retrieval fees to avoid.

1. Extract the two constraints: 120-day retention and retrieval tolerance of several hours.
2. Veto any class whose minimum duration exceeds 120 days, which eliminates Deep Archive at 180 days.
3. Among survivors, prefer the cheapest storage whose restore model fits the hours-long tolerance: Glacier Flexible Retrieval.
4. Confirm access is occasional, so retrieval fees stay negligible.
5. Attach a lifecycle expiration action at 120 days to complete the design.

Exam tip: Check retention against minimum durations before storage price; a cheaper class that outlives the data's retention is a cost trap.

Glacier Instant Retrieval keeps millisecond access at the lowest storage price that still allows it, targeting data accessed about once a quarter, with a 90-day minimum storage duration and a retrieval fee.

The requirements are quarterly access, millisecond retrieval, and seven-year retention. Millisecond retrieval immediately vetoes both asynchronous Glacier tiers: Flexible Retrieval restores in minutes to hours and Deep Archive in about 12 hours, so their lower storage prices are irrelevant, since a cheaper option that breaks a stated requirement is wrong. Between the two synchronous survivors, Glacier Instant Retrieval stores data cheaper than Standard-IA and its retrieval fees are priced for exactly this quarterly pattern, and the 90-day minimum duration is trivially satisfied by seven-year retention. Standard-IA over-pays on storage for an access frequency lower than the monthly pattern it targets.

1. Apply the retrieval-tolerance veto first: milliseconds are required, eliminating Glacier Flexible Retrieval and Deep Archive.
2. Compare the remaining synchronous classes on the stated frequency: quarterly access matches Glacier Instant Retrieval; monthly access would match Standard-IA.
3. Verify the minimum-duration trap: seven years far exceeds the 90-day minimum, so no penalty applies.
4. Choose Glacier Instant Retrieval as the cheapest class meeting every stated requirement.

Exam tip: When milliseconds are required, only Standard, the IA classes, and Glacier Instant Retrieval qualify; quarterly frequency selects Glacier Instant Retrieval.

Cost-optimal class selection can mean moving data down to a cheaper class or moving it back up: minimum-duration and retrieval-fee traps make an infrequent-access class the expensive choice for short-lived or frequently read data.

The compliance reports are the textbook Standard-IA fit: monthly access, immediate reads required, retained long past the 30-day minimum, so transitioning them saves real money. The temporary files are in the wrong direction entirely: at 7-day retention, Standard-IA's 30-day minimum bills 23 unused days per object, so returning them to Standard, which has no minimum, is a cost reduction. The working dataset must stay in Standard because daily retrieval fees in any IA class would exceed the storage savings. Deep Archive for the reports fails the immediate-access requirement, and One Zone-IA for the temp files keeps the same 30-day minimum-duration trap while adding single-AZ risk.

1. Classify each dataset by access frequency and retention: daily access, monthly access with immediate reads, and 7-day-lifetime files.
2. Match monthly-plus-immediate to Standard-IA and transition the reports.
3. Check the temp files against IA's 30-day minimum: 7-day retention fails it, so move them back to Standard.
4. Reject IA for the daily dataset by pricing the retrieval fees mentally.
5. Reject Deep Archive wherever immediate access is stated.

Exam tip: IA classes save money only when data is both long-lived and rarely read; otherwise Standard is the cheaper class.

S3 Intelligent-Tiering is the class for unknown or changing access patterns: it monitors each object and moves it between frequent, infrequent, and archive-instant tiers automatically, charging a small per-object monitoring fee but no retrieval fees and no minimum-duration charges.

The scenario states the two Intelligent-Tiering passwords: unpredictable access patterns and no operational overhead. Intelligent-Tiering adapts per object in both directions, moving reheated objects back to the frequent tier, something no lifecycle rule can do, and its lack of retrieval fees means a wrong guess about a pattern never generates a surprise bill. Standard-IA gambles that everything is cold and loses through retrieval fees on hot objects. The lifecycle option encodes a guess about cooling that the scenario says nobody can make, and it is one-way. The manual-review option fails the no-operational-effort requirement directly and reacts slowly.

1. Spot the trigger phrases: unknown or unpredictable access patterns plus minimal operational effort.
2. Eliminate hand-picked IA and Glacier classes, which require a known infrequent pattern to pay off.
3. Eliminate lifecycle waterfalls, which encode a predicted cooling curve and never move objects back up.
4. Choose Intelligent-Tiering and accept the small monitoring fee as the price of automatic adaptation.

Exam tip: Unknown or changing access patterns plus no ops overhead selects S3 Intelligent-Tiering.

Intelligent-Tiering handles per-object unpredictable patterns in both directions, but it does not monitor very small objects, so tiny records must be aggregated before any tiering strategy can help them.

The videos show the exact pattern Intelligent-Tiering exists for: some objects hot for years, others cold quickly but occasionally reheating, which a one-way lifecycle waterfall handles badly and a hand-picked IA class punishes with retrieval fees. The small metadata records are the second, separate problem: Intelligent-Tiering leaves very small objects at the frequent-access rate and IA classes bill a minimum object size, so the fix is batching them into larger objects, which also reduces billed PUT requests. The lifecycle-to-IA and Glacier Instant Retrieval options both charge retrieval fees on popular content, and One Zone-IA fails on both the small-object economics and single-AZ durability for a sole copy.

1. Identify the video access pattern as unpredictable and bidirectional, selecting Intelligent-Tiering.
2. Reject one-way lifecycle transitions and hand-picked infrequent classes because reheating content triggers retrieval fees.
3. Recognize that very small objects are not monitored by Intelligent-Tiering and gain little from IA minimum billable sizes.
4. Aggregate the small records into larger objects before upload to enable tiering and cut request charges.

Exam tip: Intelligent-Tiering for unpredictable objects, and batch tiny objects into bigger ones so any tiering can work at all.

A lifecycle waterfall assigns each phase of an object's life to the cheapest class that meets that phase's access and durability requirements, then verifies every transition against minimum durations and every archive tier against the stated retrieval tolerance.

Phase one is frequent access, so Standard avoids retrieval fees. Phase two needs immediate opens at occasional frequency, which is Standard-IA's exact pattern, and the day-30 transition satisfies its 30-day minimum. Phase three tolerates 12-hour retrievals, which is precisely Deep Archive's restore profile, and six remaining years clear its 180-day minimum many times over, making the cheapest storage class on S3 the correct archive tier. The expiration action enforces the 7-year deletion. Option a bleeds retrieval fees in month one. Option b pays Glacier Flexible Retrieval's higher storage rate to buy restore speed no requirement asks for. Option c saves a little with One Zone-IA but customer statements are a sole authoritative copy, so single-AZ storage silently breaks durability.

1. Map the three stated phases: frequent for 30 days, occasional-but-immediate to 1 year, archival with 12-hour tolerance to year 7.
2. Assign Standard, then Standard-IA at day 30, matching IA's minimum duration and transition timing.
3. For the archive phase, match the 12-hour tolerance to Deep Archive and confirm 6 years clears the 180-day minimum.
4. Add an expiration action at 7 years to enforce deletion.
5. Veto options that put hot data in IA, pay for unneeded restore speed, or store the sole copy in one AZ.

Exam tip: Build the waterfall phase by phase, then let retrieval tolerance and minimum durations veto the pretenders.