EFS vs FSx: Windows, Lustre, NetApp ONTAP or OpenZFS

Try 3 questions on this

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

1. A company runs NetApp ONTAP on premises and wants to move a file-based application to AWS without changing application code or retraining its storage team, while keeping an on-premises cache of the working set during the transition. Which approach will meet these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

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

   Your answerA.Run ONTAP software on EC2 instances so that the team keeps its tooling and manages the file servers itself.Self-managing file servers restores the provisioning, patching and failover a managed service removes.Your answerB.Use Amazon EFS and have the storage team learn its interfaces, since NFS is the protocol the application uses.EFS is a different file system; the requirement is explicitly to keep the code and the operating model.Your answerC.Use S3 with a file interface, since object storage removes the ONTAP skills.Object storage changes both the access pattern and the team's tooling, which the requirement rules out.The answerD.Use FSx for ONTAP, which keeps the familiar ONTAP features and supports SnapMirror replication and FlexCache.The migration case is the one the service is described for. "For customers who use NetApp ONTAP on-premises, FSx for ONTAP is an ideal solution to migrate, back up, or burst your file-based applications from on-premises to AWS without the need to change your application code or how you manage your data." The two named features cover the transition itself: "support for NetApp's SnapMirror replication feature" and "support for NetApp's FlexCache on-premises caching solution."

   Written from docs.aws.amazon.com/fsx/latest/ONTAPGuide/what-is-fsx-ontap.html

2. A rendering farm mounts a Regional Amazon EFS file system from hundreds of clients and the workload is highly parallel. An engineer proposes Max I/O performance mode with Elastic throughput to raise aggregate IOPS. Per-operation latency matters to the render workers, and the file system must keep scaling throughput automatically. Which approach will meet these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

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

   Your answerA.Create two file systems in Max I/O mode and shard the render assets across them to double the available IOPS ceiling.Answers a ceiling that General Purpose with Elastic throughput does not impose, and splits one namespace into two for the render workers to reconcile.Your answerB.Create the file system with Max I/O performance mode and Bursting throughput, and monitor the PercentIOLimit metric for headroom.Takes the latency penalty and gives up the automatic scaling the requirement names, because Max I/O cannot be combined with Elastic throughput.Your answerC.Keep General Purpose performance mode and switch to Provisioned throughput sized to the farm's measured peak render throughput.Provisioned throughput is the right choice when the requirement is known and the average-to-peak ratio is 5% or more.The answerD.Keep General Purpose performance mode with Elastic throughput, because Max I/O is not supported with Elastic and costs latency.The proposal cannot be built and would not help. AWS states that "Max I/O mode is a previous generation performance type", that it "is not supported for One Zone file systems or file systems that use Elastic throughput", and, twice on the page, that "for faster performance, we recommend always using General Purpose performance mode" because of Max I/O's higher per-operation latencies. The IOPS the farm needs are available without it: a Regional file system with General Purpose and Elastic throughput is specified at 900,000 to 2,500,000 read IOPS and 500,000 write IOPS, against 35,000 and 7,000 for the Bursting configuration the Max I/O proposal would force.

   Written from docs.aws.amazon.com/efs/latest/ug/performance.html

3. A genomics pipeline mounts Amazon FSx for Lustre linked to an Amazon S3 bucket and writes results into the file system. The team wants every result written by the pipeline to reach the bucket as it is produced, and wants the file system to be deletable afterwards without losing the results. Which approach will meet these requirements?

   Right. In the game that’s Renown for your Design High-Performing Architectures building.

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

   Your answerA.Rely on the linked data repository, because a file written to a file system with an association is stored in the bucket rather than on the file system.Objects become files through the association, and file contents are lazily loaded on first access. Writes are a separate mechanism with its own setting.Your answerB.Run an export data repository task on a schedule and turn on automatic export as well, so that nothing written between tasks is missed.AWS is explicit that automatic export and export data repository tasks can't be used simultaneously on a file system. Import is the direction where both may run together.Your answerC.Take a backup of the file system after each pipeline run so that the results are retained independently of the linked S3 bucket.Backups are not supported on a file system linked to an S3 data repository, so this is unavailable here as well as slower.The answerD.Enable automatic export on the data repository association, exporting file data and metadata as files are created and modified.Automatic export is the documented continuous direction. AWS states that "when you turn on automatic export on a data repository association, your file system automatically exports file data and metadata as files are created, modified, or deleted", which satisfies both halves of the requirement: results reach the bucket as they are produced, and the durable copy survives the file system. The alternative, an export data repository task, exports "file data and metadata that were created or modified since the last such task", which is a batch rather than a stream.

   Written from docs.aws.amazon.com/fsx/latest/LustreGuide/fsx-data-repositories.html