Design Cost-Optimized Architectures practice questions

Answer: A) An Elastic IP address that is still allocated to the account but isn't associated with any resource; B) Amazon EBS volumes left in the available state after their instances were terminated, and their old snapshots

Elastic IP addresses are charged whether or not they are associated, and EBS volumes and snapshots are billed until deleted. A stopped instance's compute, an empty Auto Scaling group, and IAM roles or security groups add no charges.

• ACorrect. AWS charges for every Elastic IP address in the account, whether it is associated or not. An unassociated one is pure waste until it is released.
• BCorrect. EBS volume storage is billed for the provisioned GB until the volume is deleted, and snapshot storage is billed for the data it holds, whether or not anything uses them.
• CIncorrect. You aren't charged for instance usage or data transfer while an instance is stopped. Only its EBS volumes, plus any associated Elastic IP, keep incurring charges.
• DIncorrect. EC2 Auto Scaling has no additional fees beyond the EC2 instances and other resources it uses. A group with no instances incurs no charges.
• EIncorrect. IAM roles and security groups are offered at no additional charge, so leftover ones are a security-hygiene issue rather than a cost issue.

Answer: A) Use a schedule to stop the instances outside business hours and start them each weekday morning

For workloads needed only during business hours, scheduled stop/start removes most billable instance hours. EBS storage keeps billing while instances are stopped. Long-term commitments sized for 24/7 usage still pay for the idle hours.

• AStopped instances aren't billed for instance usage, though attached EBS volumes are still charged. Cutting runtime from 168 to about 50 hours a week saves up to roughly 70% of compute cost. Instance Scheduler on AWS, or EventBridge with Lambda, can automate this without architectural changes.
• BA Savings Plan lowers the hourly rate but commits the company to paying for every hour of the term, including the roughly 70% of hours when the environment isn't needed.
• CWith a fixed desired capacity, the group keeps the same number of instances running around the clock. It improves availability but doesn't remove idle hours.
• DDetailed monitoring adds CloudWatch charges and only provides data. By itself it doesn't reduce the compute bill.

Answer: A) Cover the steady baseline with a Compute Savings Plan or Reserved Instances and handle bursts with On-Demand or Spot

• ACommitting to the always-on portion captures the biggest discount, while flexible capacity handles the variable part.
• BPaying full On-Demand rates for a predictable 24/7 baseline leaves substantial savings on the table.
• CReserving to the peak means paying for capacity that sits idle most of the day.
• DUnhandled Spot interruptions would compromise a production application's availability.

Answer: A) Lifecycle rule: S3 Standard-IA at 30 days, S3 Glacier Deep Archive at 365 days, expire at 2,555 days

When access patterns are predictable, use a lifecycle policy that matches each phase to the cheapest class that meets its access and resilience needs. Here that is Standard-IA for millisecond monthly access across multiple AZs, then Glacier Deep Archive for rarely read data, with expiration at seven years.

• ACorrect. Standard-IA is designed for data accessed about once a month that still needs millisecond access, and it stores data across multiple Availability Zones. Deep Archive is the lowest-cost class for data read less than once a year, and its restores take hours. Expiration removes the objects after seven years.
• BIncorrect. Objects in Glacier Flexible Retrieval are archived and must be restored before they can be read. That breaks the requirement for millisecond reads during the first year.
• CIncorrect. One Zone-IA stores data in a single Availability Zone and isn't resilient to the loss of that zone. AWS recommends Standard-IA for a primary copy that can't be re-created.
• DIncorrect. Intelligent-Tiering is best for unknown or changing access patterns and charges a per-object monitoring fee. This access pattern is predictable. Without the optional archive tiers, the data would also stay in Archive Instant Access for years instead of moving to the cheaper Deep Archive class.

Answer: A) An S3 Lifecycle policy transitioning objects Standard → Standard-IA at 30 days → Glacier Flexible Retrieval at 90 days → Glacier Deep Archive at 365 days

• ALifecycle transitions match storage cost to the declining access pattern automatically, and Deep Archive keeps long-tail data retrievable at the lowest price.
• BStandard is far more expensive than colder classes for data that is rarely or never accessed after the first months.
• CEBS is block storage priced well above S3 archival classes and isn't designed for seven-year retention of cold logs.
• DThat violates the seven-year compliance requirement.

Answer: A) A gateway VPC endpoint for S3

• AA gateway endpoint routes S3 traffic privately within the AWS network rather than through the NAT gateway, eliminating the NAT gateway's per-gigabyte data processing charge for that traffic entirely, at no additional hourly cost for the endpoint itself.
• BNAT gateways don't have a user-selectable 'size' in the way EC2 instances do, and this wouldn't reduce the per-gigabyte data processing charge regardless.
• CA private subnet by definition has no direct route to the internet without a NAT gateway or similar — and this would also change the subnet's security posture, which the requirement says to avoid.
• DThis removes required application functionality rather than optimizing the cost of it.

Answer: A) Move the fleet to Graviton-based instances for better price-performance

• AGraviton (ARM) instances frequently cost less for equivalent throughput on scale-out and CPU-bound workloads, and a JVM service typically runs on ARM with no code change.
• BThat increases the rate paid, not decreases it.
• CAdding capacity raises cost; it doesn't improve price-performance per unit of work.
• DThat slightly increases cost and doesn't reduce compute spend.

Answer: A) Use scheduled automation to stop the DB instances each evening and start them again each weekday morning

For non-production RDS databases used only part of the time, stopping and starting them on a schedule removes instance-hour charges while they are idle. Storage still bills and the data is kept. Reservations and extra replicas do not fit part-time use.

• AWhile an RDS DB instance is stopped, you are not charged for DB instance hours. Provisioned storage and backup storage still bill, and the data is kept. A weekday-only schedule removes the compute charge for nights and weekends, which is roughly 70% of the week. The instance's endpoint and configuration stay the same. RDS auto-starts an instance after 7 consecutive stopped days, but a daily schedule never reaches that limit.
• BA Reserved DB Instance is billed for every hour of the term whether or not the database is running. For databases needed only about 50 hours a week, stopping them outside business hours saves more than a reservation discount, and stopping a reserved instance would waste the commitment.
• CRDS does not let you stop a DB instance that has a read replica or that is a read replica. The replicas would also add instance-hour charges instead of removing them.
• DMulti-AZ adds a standby that roughly doubles instance cost. You cannot stop only the standby: a stop action applies to the whole DB instance deployment.

Answer: A) Spot Instances

• ASpot Instances offer the deepest discount versus On-Demand in exchange for the possibility of interruption — ideal for a fault-tolerant batch job that can simply retry or resume a worker if one is reclaimed.
• BA long-term reservation makes sense for steady, predictable, always-on workloads — it doesn't specifically minimize cost for a flexible, interruption-tolerant nightly batch job the way Spot does.
• COn-Demand is more expensive than Spot for exactly the kind of interruption-tolerant workload described, leaving savings unused.
• DDedicated Hosts are priced for dedicated physical capacity tied to licensing/compliance needs, not for minimizing cost on a flexible, interruptible batch job.

Answer: A) Run the batch tasks with the Fargate Spot capacity provider in the ECS cluster

Fargate Spot offers discounted Fargate capacity for interruption-tolerant tasks, with a two-minute warning before a task is stopped. That makes it the right fit for batch jobs that can be restarted.

• ACorrect. Fargate Spot runs interruption-tolerant tasks on spare capacity at a discount compared with the Fargate price. Tasks can be interrupted with a two-minute warning, which is acceptable because these jobs checkpoint and restart.
• BIncorrect. Bigger tasks cost more per hour on regular Fargate, and nothing says the jobs would finish fast enough to make up for it. This does not reliably deliver significant savings.
• CIncorrect. EC2 Reserved Instances discount EC2 instance usage, not Fargate tasks. They also lock in capacity that intermittent batch jobs would leave mostly idle.
• DIncorrect. Dedicated Instances cost more than shared-tenancy capacity and add server management, which works against both the cost goal and the serverless requirement.