SAA-C03 Practice Question 87: Instance Store Scratch Space
AWS Certified Solutions Architect – Associate (SAA-C03) · Test 9 · question 7 of 10
A video transcoding fleet writes very large intermediate files while processing each job. The scratch directory needs the highest possible local IOPS and throughput at the lowest cost, the files are worthless once the job finishes, and any instance that dies simply has its job requeued and processed from the original in S3.
Which storage should the intermediate files use?
Show answer and explanation
Answer:
An instance store is temporary block-level storage on disks physically attached to the host, and AWS states it is ideal for temporary storage of information that changes frequently, such as buffers, caches and scratch data. That is exactly what these intermediate files are. Because the disks are local rather than network-attached, they deliver the highest local IOPS and throughput of the options, and there is no additional charge; instance store volumes are included in the usage cost of the instance, so the cost requirement is met too. The usual objection, that the data is lost when the instance stops or terminates, does not apply: the job is simply requeued and rerun from the source object in S3.
Option A is a shared network file system. It adds network latency per operation, charges for storage and throughput, and shares scratch space across instances that have no reason to see each other's temporary files.
Option B is network-attached block storage. Provisioning gp3 to its ceiling costs money for every instance and still tops out well below what local NVMe delivers, all to protect data that nobody wants to keep.
Option C compounds that. io2 Block Express is the premium durable volume type, and snapshotting scratch files after each job pays to store data that is defined as worthless the moment the job ends.
Reference: Instance store temporary block storage for EC2 instances