Cost and Performance Optimization practice questions

Answer: A) Provisioned concurrency of 50 on the function's published alias

Provisioned concurrency keeps a set number of execution environments pre-initialized on a version or alias, which removes cold starts for that capacity at extra cost. Reserved concurrency only reserves and caps concurrency, and SnapStart restores environments on demand instead of keeping them warm.

• ACorrect. Provisioned concurrency allocates pre-initialized execution environments that are ready to respond immediately, which removes cold starts for that capacity. It must be set on a version or alias, not on $LATEST, and it incurs additional charges, which matches the team's trade-off.
• BIncorrect. Reserved concurrency sets aside part of the account's concurrency for the function and caps how far it can scale. It does not pre-initialize execution environments, so cold starts still happen. It also has no additional charge, so it isn't the paid warm-capacity feature the team is asking for.
• CIncorrect. SnapStart does not keep environments initialized ahead of requests. It resumes new environments on demand from a cached snapshot to shorten startup, and AWS documents that it can't be used together with provisioned concurrency. Also, SnapStart currently supports only Java, Python and .NET managed runtimes, not Node.js.
• DIncorrect. More memory also gives the function more CPU and can shorten initialization somewhat, but Lambda still has to create and initialize new environments on demand. Cold starts remain.

Answer: A) Opt in to AWS Compute Optimizer and review its EC2 instance recommendations

AWS Compute Optimizer makes EC2 rightsizing recommendations from observed CloudWatch utilization metrics. It includes memory utilization when the CloudWatch agent publishes memory metrics.

• ACorrect. After opt-in, Compute Optimizer analyzes the instances' specifications and CloudWatch utilization metrics (14 days by default) and recommends rightsized instance types. It includes memory utilization because the CloudWatch agent is publishing memory metrics.
• BIncorrect. Target tracking in EC2 Auto Scaling changes the number of instances to keep a metric at a target value. It doesn't recommend a different instance type or size.
• CIncorrect. Savings Plans recommendations suggest a commitment to buy based on past usage. They lower the rate paid for existing usage, but they don't identify over-provisioned instances or suggest smaller instance types.
• DIncorrect. Anomaly detection alarms alert when a metric moves outside its expected band. They don't analyze utilization against instance specifications or recommend a different instance type.

Answer: A) An S3 Lifecycle rule with the AbortIncompleteMultipartUpload action and a DaysAfterInitiation value

AWS recommends a lifecycle rule with the AbortIncompleteMultipartUpload action. After a set number of days, it aborts incomplete multipart uploads and deletes their stored parts. Expiration rules and noncurrent-version rules act on objects, not on upload parts.

• ACorrect. If a multipart upload is not completed within the set number of days after it starts, the AbortIncompleteMultipartUpload action makes it eligible for abort. Amazon S3 then stops the upload and deletes its parts. The rule applies to existing uploads and to future ones, and it has no effect on completed objects.
• BIncorrect. Expiration applies to completed objects, so this rule would delete real application data. It does not target the parts left behind by unfinished multipart uploads.
• CIncorrect. NoncurrentVersionExpiration removes older versions of objects in a versioned bucket. Parts from incomplete multipart uploads are not object versions, so this rule does not remove them.
• DIncorrect. Transfer Acceleration can make long-distance uploads faster, but it adds cost. It does not abort uploads that were abandoned or delete parts that already exist.

Answer: A) A mixed instances group that lists the eligible instance types and uses a price-based allocation strategy such as lowest-price

To diversify across instance types and let AWS choose by price, use a mixed instances group with several instance types (or attribute-based selection) and an allocation strategy such as lowest-price or price-capacity-optimized. Scaling policies and launch template versions do not select instance types by price.

• ACorrect. A mixed instances group lets you list several instance types, or use attribute-based instance type selection. An allocation strategy then controls which types are launched. The On-Demand lowest-price strategy launches the lowest-priced type in each Availability Zone. For Spot, price-capacity-optimized weighs both price and available capacity.
• BIncorrect. Scaling policies change how many instances the group runs (its desired capacity). They do not decide which instance types are launched. Only the group's instance type configuration does that.
• CIncorrect. An Auto Scaling group launches from the one launch template version it is configured to use. It does not rotate between versions based on price. Choosing among several types by price requires a mixed instances group with an allocation strategy.
• DIncorrect. With only one instance type set, the group can launch nothing but m5.large. Capacity rebalancing helps replace Spot Instances at risk of interruption. It does not add instance type variety or pick types by price.

Answer: A) Define a table over the S3 prefix and query the needed columns with Amazon Athena

Amazon Athena runs standard SQL directly against data in Amazon S3 without servers to manage, so the job no longer has to download and parse whole CSV files. S3 Select is no longer available to new customers, so Athena is the current choice for this pattern.

• ACorrect. Athena is a serverless query service that analyzes data directly in Amazon S3 using standard SQL, and you pay only for the queries you run. The job gets back only the query results, not the whole CSV files. Converting the data to a columnar format such as Parquet can reduce the data each query scans even more.
• BIncorrect. Transfer Acceleration can speed up long-distance transfers to and from a bucket, but the job still downloads and parses every full object, and acceleration adds its own transfer charge.
• CIncorrect. Requester Pays only changes who pays for requests and data transfer. The same full objects are still downloaded and processed, so nothing is saved.
• DIncorrect. Moving objects to Standard-IA lowers storage cost but adds per-GB retrieval charges. The job would still download and parse the entire files nightly, which could cost more.

Answer: A) Sign in to the management account, open Billing preferences, and deactivate RI and Savings Plans discount sharing for that member account

The management account controls RI and Savings Plans discount sharing. It can deactivate sharing for specific member accounts in Billing preferences, and member accounts can't change this themselves.

• ACorrect. The management account of an organization controls RI and Savings Plans discount sharing. In Billing preferences, under the discount sharing preference by account, it can deactivate sharing for individual member accounts.
• BIncorrect. Discount sharing preferences are controlled by the organization's management account, not by individual member accounts. A member account can't opt itself out.
• CIncorrect. SCPs limit the IAM permissions available in member accounts. They do not control how billing discounts are applied across a consolidated bill.
• DIncorrect. Organizational units don't define discount sharing boundaries. Sharing is set in the management account's Billing preferences, either per account or through sharing groups defined with Cost Categories.

Answer: A) Create an Intelligent-Tiering archive configuration on the bucket that activates the Archive Access and Deep Archive Access tiers

Intelligent-Tiering moves objects through its Frequent, Infrequent, and Archive Instant Access tiers automatically. The cheaper Archive Access and Deep Archive Access tiers are used only after you activate them in an archive configuration.

• ACorrect. The Archive Access tier (default threshold 90 days with no access) and the Deep Archive Access tier (default 180 days) are optional and must be activated. Once active, Intelligent-Tiering moves objects into them automatically, at costs comparable to S3 Glacier Flexible Retrieval and S3 Glacier Deep Archive.
• BIncorrect. The Archive Instant Access tier is automatic after 90 days with no access and offers millisecond retrieval. It doesn't move objects into Deep Archive-level storage. The Deep Archive Access tier is only used if you activate it.
• CIncorrect. S3 Storage Lens gives visibility and recommendations about storage usage. It doesn't control or activate the Intelligent-Tiering archive tiers.
• DIncorrect. This takes the objects out of Intelligent-Tiering, which the company wants to avoid. One Zone-IA is also an infrequent-access class, not an archive tier, and it stores data in a single AZ.

Answer: A) Compute Savings Plans with an hourly spend commitment

Compute Savings Plans are the most flexible Savings Plan type. They apply across EC2 instance families, sizes and Regions, and also cover Fargate and Lambda.

• ACorrect. Compute Savings Plans apply to EC2 usage regardless of instance family, size, Region, operating system or tenancy. They also apply to Fargate and Lambda usage, so the discount continues through every planned change.
• BIncorrect. EC2 Instance Savings Plans require a commitment to one instance family in a chosen Region, such as m5 in one Region. They do not follow usage that moves to another family, another Region, Fargate or Lambda.
• CIncorrect. Convertible Reserved Instances can be exchanged for other EC2 instance configurations, but they only discount EC2 instances. They do not apply to Fargate or Lambda usage.
• DIncorrect. Standard Reserved Instances are tied to specific instance attributes and cannot be exchanged for a different instance family. They also only discount EC2, not Fargate or Lambda.

Answer: A) A target tracking scaling policy that keeps the group's average CPU utilization at 50%

Target tracking scaling adjusts Auto Scaling capacity in response to the actual value of a metric such as average CPU utilization, so it handles unpredictable load in both directions. Scheduled and predictive scaling depend on predictable or recurring patterns.

• ACorrect. Target tracking is a dynamic scaling policy. It adds and removes capacity based on a CloudWatch metric and a target value, so the group scales in proportion to the actual load. Because it reacts to observed load, it does not need the load to follow a predictable pattern.
• BIncorrect. Scheduled actions change capacity at fixed times. When load does not follow a predictable schedule, the group will be too large or too small for much of the day.
• CIncorrect. Predictive scaling looks for daily or weekly patterns in historical load and launches capacity ahead of the forecast. It suits cyclical or recurring workloads. This workload has no recurring pattern to forecast from.
• DIncorrect. This keeps the group at peak size all the time. Performance is maintained, but the group never scales in during quiet periods, so it does not reduce cost.

Answer: A) A Compute Savings Plan

Compute Savings Plans are the most flexible commitment. They apply automatically to EC2 usage regardless of instance family or Region, and to Fargate and Lambda usage. EC2 Instance Savings Plans and Reserved Instances cover EC2 only.

• ACorrect. Compute Savings Plans apply to EC2 instance usage in any instance family, size, Region, OS or tenancy. They also apply to Fargate and Lambda usage.
• BIncorrect. An EC2 Instance Savings Plan is a commitment to one instance family in one Region. It doesn't apply to Lambda or Fargate usage.
• CIncorrect. Standard Reserved Instances discount matching EC2 instance usage only. They don't apply to Lambda or Fargate.
• DIncorrect. Convertible Reserved Instances can be exchanged for other EC2 instance configurations, but they still cover only EC2 instances, not Lambda or Fargate.