search

daita-technologies / backend

Backend system for the DAITA platform.
http://app.daita.tech
GNU Affero General Public License v3.0
3 stars 1 forks source link

Implement AWS Fargate #6

Open pcaversaccio opened 2 years ago

pcaversaccio commented 2 years ago

As part of the architecture refactorisation, we should start using AWS Fargate. If we use a serverless, pay-as-you-go compute engine we can optimise memory management as well as our cost strategy.

ttattl commented 2 years ago

Currently, AWS Fargate does not support GPU (CPU is ok). https://github.com/aws/containers-roadmap/issues/88 https://aws.amazon.com/blogs/architecture/building-real-time-ai-with-aws-fargate/ I think we can keep our structure until we get full support from Fargate. We can see some recommendations in Git but basically, it looks the same with us right now. So, let's wait for the full version of AWS. @giaoNguyen70 and @pcaversaccio how do you think ?

pcaversaccio commented 2 years ago

Hmm but wouldn't be Fargate an improvement for our EC2 CPU instance management still? I agree on the GPU part and we can wait for that but AWS Fargate could be also relevant for the other non-GPU-related processing.

ttattl commented 2 years ago

I am considering keeping the current flow or changing to Fargate. Currently, our flow also works similarly to Fargate, and also can support GPU if we need it. If we change now I am not sure about GPU.

pcaversaccio commented 2 years ago

Let's re-evaluate the AWS Fargate issue (in particularly re GPUs) in the next phase again (i.e. May 2022). However, for the moment we go with our existing, custom-built architecture.