I am considering migrating my AWS Batch workflow from x86 to graviton(arm64). We use fusion/wave for the file system and docker/ECR for containers. The wave performance is a pain point which we mitigate with wave.freeze.
The plan is to build natively on both x86 and arm64 and weave them together with the docker manifest to create a multi-architecture image. [as described in this video Migrating Containers to AWS Graviton - AWS. The presenter starts talking about this topic at around 10min 21 secs.]
Given this combined docker image, is this going to adversely effect wave performance?
Wave runs in any case a separate build depending your are requesting an AMD or ARM architecture, therefore there should not be a specific impact on performance assuming you were referring longer build time for multi-arch container images.
Said best practice consist using pre-build ARM-enabled Bioconda packages which are generally more reliable and quicker to build compared to Dockerfile.