fault tolerant migration from KVM to EC2

Question

I am working on a project in my university, and it's considering the feasibility of create a hybrid infrastructure between the on-premise data center (it's a very small server room with 2 racks almost fully loaded) to AWS.

I am wondering if there exist a way to migrate VMs based on KVM (we are using RHEV in the on-premise) in way that VMs can be instantaneously deployed in EC2 if there is a disaster in the on-premise (UPS or CRAC units failures, network failures, or natural disasters).

Sadly, Server Migration Service in AWS only supports VMWare at this time, and moving directly RAW images requires the VMs on the on-premise to be stopped for fully consistency (most of the times this will not be possible). I will not be using direct connect this time, so the process of moving every VM will be kind of slow.

I've thought of creating an EFS shared between a VPC and the on-premise, and constantly updating raw images in EFS, but I don't know if there are better approaches.

Answer 1

Interesting question.

the feasibility of create a hybrid infrastructure between the on-premise data center [..] to AWS

This is definitely possible, but it doesn't sound like you actually want a hybrid environment. It sounds like you want to directly run you KVM images as if they AMI's, which i'm not convinced is a good idea.

Are you using any kind of CM system for your on-premise images? Or are they "hand tuned" configurations? I would suggest that you configure parallel instance images (AMIs) on AWS.

in way that VMs can be instantaneously deployed in EC2 if there is a disaster

Nothing is "instantaneous" ;)

The hard part of a hybrid infrastructure is your data. Your data storage layers, and keeping them synced between local and cloud is the modern challenge. There are loads of interesting new technologies being developed to solve exactly that problem. I would suggest you figure out a good replication strategy between your local and cloud datastores. Trying to replicate machine images is probably not the useful level of abstraction.