Lambda vs Fargate vs EC2: choosing compute by cost and workload

Try 3 questions on this

Pick an answer to see why each option is right or wrong.

1. A Lambda function downloads 4 GB satellite images from Amazon S3 and processes them. It currently holds each image in memory, so memory is provisioned far above what the computation needs, and the per-millisecond cost reflects that. Which solution meets these requirements?

   Right. In the game that’s Renown for your Design Cost-Optimized Architectures building.

   Not that one: D is right. In the game your Design Cost-Optimized Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.Keep the memory setting and add a smaller second function that streams the image through in fragments.Two functions is more moving parts than a storage setting, and the fragments still have to be reassembled somewhere.Your answerB.Move the workload to an EC2 instance instead, since Lambda's 512 MB of temporary storage cannot hold files of this size.Ephemeral storage goes up to 10,240 MB, so the file fits without leaving Lambda.Your answerC.Attach an EFS file system to the function and write the image there, because /tmp cannot be resized.EFS is for shared or persistent data; /tmp is resizable and is the documented fit for scratch space.The answerD.Increase ephemeral storage, write the image to /tmp, and set memory to what the computation needs.The scratch space is a separate dial from memory. Ephemeral storage is configurable "between 512 MB and 10,240 MB, in 1-MB increments", and the documented use cases name this workload directly: "for workloads that download objects from Amazon S3 in response to S3 events, more /tmp space makes it possible to handle larger objects without using in-memory processing", with graphics processing called out for "large TIFF files or satellite images". Sizing memory for the computation and /tmp for the file is what separates the two costs.

   Written from docs.aws.amazon.com/lambda/latest/dg/configuration-ephemeral-storage.html

2. A company runs an interruption-tolerant Linux batch service on Amazon ECS entirely on the FARGATE_SPOT capacity provider, with all tasks in subnets of one Availability Zone. Tasks often fail to start or stop early with SpotInterruption errors. The company must keep every task on Fargate Spot pricing, must give the service more Spot capacity to draw on, and must record every interruption for later review. Which combination of steps should a solutions architect take to meet these requirements? (Select TWO.)

   Right. In the game that’s Renown for your Design Cost-Optimized Architectures building.

   Not that one: C,E is right. In the game your Design Cost-Optimized Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.Add the FARGATE capacity provider to the strategy with a weight of 1 so that regular Fargate takes over whenever Spot capacity runs out.Tasks placed on the FARGATE provider are billed at the regular Fargate rate, which breaks the Spot-pricing requirement, and Fargate does not replace Spot capacity with regular capacity on its own.Your answerB.Raise the stopTimeout in the container definition to 300 seconds so that interrupted tasks keep running until their work finishes.The warning before a Spot task is stopped is two minutes, and the documented stopTimeout for this is 120 seconds or less, so 300 seconds does not buy extra running time.Your answerC.Add subnets in other Availability Zones to the service's network configuration so that its tasks can run in several Availability Zones.Your answerD.Increase the service's desired count so that more tasks are started in the same Availability Zone whenever Spot capacity is reclaimed.More tasks in the same Availability Zone draw on the same Spot capacity. The documented remedy for SpotInterruption errors is running tasks in more Availability Zones.Your answerE.Create an Amazon EventBridge rule that matches ECS Task State Change events with the SpotInterruption stop code and sends them to a log target.

   Show the full explanation and doc quotes

   That is C. A record of every interruption: "When tasks that use the Fargate and Fargate Spot capacity providers are stopped, the task state change event is sent to Amazon EventBridge. The stopped reason describes the cause", and a Spot interruption carries "stopCode": "SpotInterruption". A rule matching those events (E) records each one. C alone leaves no record, and E alone records interruptions without reducing them.

   Written from docs.aws.amazon.com/AmazonECS/latest/developerguide/spot-interruption-errors.html

3. A company runs a steady compute baseline that moves between Amazon EC2 instance families, AWS Fargate, and AWS Lambda, and it wants a discount on that baseline. During a six-week data center cutover it also needs assured capacity for 40 m6i.xlarge instances in one Availability Zone, released afterward. The discount must also apply to those 40 instances. Which combination of steps should a solutions architect take to meet these requirements? (Select TWO.)

   Right. In the game that’s Renown for your Design Cost-Optimized Architectures building.

   Not that one: B,D is right. In the game your Design Cost-Optimized Architectures building would lose a little Renown, and this question would come back later.

   Your answerA.Buy an EC2 Instance Savings Plan for m6i to cover EC2, Fargate, and Lambda.An EC2 Instance Savings Plan applies within one EC2 instance family and does not apply to Fargate or Lambda.Your answerB.Purchase a Compute Savings Plan with an hourly commitment sized to the company's steady compute baseline.Your answerC.Rely on the Compute Savings Plan to hold capacity for the 40 m6i.xlarge instances during the cutover.Savings Plans do not provide capacity reservations; the plan discounts usage but holds no capacity for the cutover.Your answerD.Create an On-Demand Capacity Reservation for 40 m6i.xlarge instances that starts immediately, and cancel it after the cutover.Your answerE.Purchase zonal Standard Reserved Instances for the 40 m6i.xlarge instances in that Availability Zone for the cutover.Reserved Instances are bought for a one-year or three-year term, so the capacity cannot be released after six weeks.

   Show the full explanation and doc quotes

   Savings Plans "provide savings beyond On-Demand rates in exchange for a commitment of using a specified amount of compute power (measured per hour) for a one or three year period", and a Compute Savings Plan also covers Fargate and Lambda. It holds no capacity: "Savings Plans doesn't provide capacity reservations, but you can allocate On-Demand Capacity Reservation (ODCR) for your needs and your Savings Plans will apply." An immediate reservation has "no term commitment" and can be cancelled "to release the reserved capacity and to stop incurring charges."

   Written from docs.aws.amazon.com/savingsplans/latest/userguide/what-is-savings-plans.html