Mats Lannér

I did a few quick experiments on SOCI support for Fargate / ECR, comparing the gains for different image sizes. tl;dr – for big images, this is a huge win. For small images it doesn't make much of a difference. Performance One thing upfront: Several places, including the quoted tweet, praise "x% faster!" or "increase in performance!" – note that this affects startup time only, so for any redundant service, you "only" get better deployment times. Cloudformation Template The blog post below included a "Partner Solution" Cloudformation stack that will take care of indexing newly pushed images. For larger images, it takes several minutes to index them, e.g. >3min for that ~2.4 GB image. You might want to consider that, because the SOCI index won't be there yet when you deploy immediately after push. Other than that, it worked flawlessly. Speed test I started a Fargate task in the same subnet as the controlling ec2 instance (1 vCPU). The duration was measured from successful `aws ecs run-task` to receiving an inbound network connect from the container (netcat from container init). 10 iterations per test. tiny image (alpine) size(docker) = 7.33 MB / size(ecr) = 3.4 MB w/ SOCI: min=11.33s avg=14.07s max=18.61s w/o SOCI: min=10.70s avg=12.98s max=15.35s => -8% faster 🙃 medium image size(docker) = 215 MB / size(ecr) = 141 MB w/ SOCI: min=12.29s avg=14.85s max=17.28s w/o SOCI: min=18.07s avg=19.90s max=22.23s => 25% faster big image size(docker) = 2.43 GB / size(ecr) = 1.0 GB w/ SOCI: min=14.43s avg=16.26s max=18.24s w/o SOCI: min=52.12s avg=65.50s max=85.88s => 75% faster So the SOCI overhead might be hurtful for tiny images, but this could just as well be a measurement problem here. For larger images, the gains are significant. SOCI index data comes at a (small) price though, as the additional entries in ECR take up some space (~6 MB for the medium image and ~34 MB for the big image). No support for Public ECR (yet?).