When you use Amazon S3 and DynamoDB for your configuration store, most services will be highly available with the following exceptions:
Why is that? What is the technical reason that GP cannot be made highly available in this scenarion? What dependency can not be met in such setup by the GP? Or is it for Custom GP tools only?
I believe the problem is that the jobs directory has to be on disk to store the scratch.gdb, pdf, and other outputs. The file server that will store the jobs directory won't be highly available as it's simply a file server. I'm not sure what the highly available storage options are in AWS, (other than S3 and DynamoDB, of course).