Samuel presented slides
We got good feedback from RH, IBM and Intel, still pending feedback from AMD
Discussing the topic of service mesh and encrypted containers - what happens when someone pushes a sidecar into the pod? What happens if that sidecar just prints all the memory to somewhere else :-)...
Idea would be to encrypt the side car as well, not guaranteeing what’s inside but yes regarding what you are getting
This does not violate our threat model which is focusing on knowing exactly what we are running
Christophe discussed the topic of the performance cost of using encrypted memory such as extra disk space - the user should decide if he wants to encrypted everything or only the top layers
Smauel pointed out that encrypting only the upper layers would not be strong enough - If your workload runs on a fedora and you only encrypted your workload, the fedora could have been hacked and added a bunch of backdoors
For IBM the goal is to work with full encryption before talking about sharing
The SEV also has a number of issues regarding density as it is so maybe the additional sharing would not save much - max 512 VMs
Note that we are talking about SEV per pod (with a kata VM behind)
We discussed virtio-fs and DAX and the fact that it can’t work with encrypted memory since you don’t want anything to be shared
We want to expand the problem statement and scope not only to fully encrypted images but also signed images