Exam AWS Certified Machine Learning - Specialty topic 1 question 196 discussion

I think the answer is A. B: Snowcone has limit with 8 TB. C: is AWS on-premises solution, but the company wants to store these transcriptions in an Amazon S3 bucket in the AWS Cloud for model development. D: 100 Mbps cannot handle petabytes datasync.

A: is the Right Folks, please... OUTPOST is very complicated to implement, and the question is not talking about continue to do it after. We know that Snowball has limite of storage, but the ideia is not to send Petabytes of data to S3, just only the likelihood processed. so, the ideia is to use the Snowball as an Optimized machine to be able to process the data and send it to S3. The Snowball Edge Compute Optimized device provides 52 vCPUs, 208 GiB of memory, and an optional NVIDIA Tesla V100 GPU. For storage, the device provides 42 TB usable HDD capacity for Amazon S3 or Amazon EBS, as well as 7.68 TB of usable NVMe SSD capacity for EBS block volumes. Snowball Edge Compute Optimized devices run Amazon EC2 sbe-c and sbe-g instances, which are equivalent to C5, M5a, G3, and P3 instances.

A. NO - Snowball needs to be shipped back to AWS, that does not use Datasynch B. YES - that is an edge computing device C. NO - Too big admin overhead to have your local AWS Cloud D. NO - Too slow over the 100Mps connection

faster

A, as this is faster option.

A ==> Bring the compute closer to the data

The key is faster Transfer speeds of up to 100gb per second

It either A or C. But C is too complicated to setup. You order rack and AWS installs that plus you need enterprise support and the biggest reason it is not possible in this case is that it requires at least 1GB connection. The question clearly asks as soon as possible so A is the best choice in my opinion

The key is to deliver the transcripts to S3 as early as possible. Outpost order and provisioning takes months. I would go for A as its logical to do local inference and send transcripts to S3.

The model needs to run on-premises to don't be necessarily upload all the audio data to after run the model and this solution can use datasync to upload the results after too, then it's a good choice!

Outpost for use case when customer don’t want to transfer their data out

It is A. C can't be the answer, as to transfer 1PB data, it may take 1,000 days under a 100 mbps network.

My vote goes to C, A. Snowball device only stores around 80 TB of data and uploading the newly transcribed data through datasync still goes through the slow connection between on-site and AWS. Outpost seem like the only feasible solution here that can satisfy both requriements

Correct Answer A. D is incorrect as it takes 1024 Days Approx to transfer Petabyte of data.

The best solution would be option A and D. Option A: Order and use an AWS Snowball Edge Compute Optimized device with NVIDIA Tesla modules to run the transcription algorithm. Use AWS DataSync to send the generated transcriptions to a transcription S3 bucket. This option allows to use a device that has the necessary GPU for running the transcription algorithm and then use the AWS DataSync to send the generated transcriptions to the S3 bucket. Option D: Use AWS DataSync to ingest the audio files to Amazon S3. Create an AWS Lambda function to run the transcription algorithm on the audio file as it is uploaded to Amazon S3. Configure the function to write the generated transcriptions to the transcriptions S3 bucket. This option allows to automatically transcribing the audio files as they are uploaded to S3. This means that the transcriptions are ready as soon as the audio files are uploaded and eliminates the need to transcribe the audio files separately.

ws.amazon.com/about-aws/whats-new/2020/07/aws-snowball-edge-compute-optimized-now-available-additional-aws-regions/

I guess C, given that "The company wants to store these transcriptions". Petabytes audio data can keep in on-prem.