I remember at the beginning Google didn't want to share their SOTA datacenter with public, considering it a market advantage. So instead of raw VMs they decided to promote the "framework" model, where all the seams are polished, under the name of Google App Engine.
I remember we've been going through a security audit, and many of their questions were about ports security and server access -- well, we did not even have SSH access at all, and only 80/443 ports were accessible. It was a breeze.
However, as market has shown, people really do want finer control, SSH root access, and custom ports. Along with many more fine-tuning. For example, GAE autoscaler did not reassign a request that triggered a new instance, waiting for it to start for 20s even though all requests completed within 200ms, so other workers were available, but the request waited for its own instance to wake up.
So I'm skeptical of "seams hidden" approach, as over and over again I needed to tune that one hidden seam.
I remember we've been going through a security audit, and many of their questions were about ports security and server access -- well, we did not even have SSH access at all, and only 80/443 ports were accessible. It was a breeze.
However, as market has shown, people really do want finer control, SSH root access, and custom ports. Along with many more fine-tuning. For example, GAE autoscaler did not reassign a request that triggered a new instance, waiting for it to start for 20s even though all requests completed within 200ms, so other workers were available, but the request waited for its own instance to wake up.
So I'm skeptical of "seams hidden" approach, as over and over again I needed to tune that one hidden seam.