Intel and AMD adding virtualization support to CPU’s marked a watershed moment in the history of virtualization. AMD-V and Intel VT closed the gap in performance between physical workloads and virtualized workloads. Hardware based support virtualization support allowed not only the support for 64-bit virtual guests but nested virtualization as well. Nested virtualization allows the running of an hypervisor within an hypervisor. A small start up, Ravello Systems hopes to leverage this engineering trick to enable a new level of capability for virtualized environments hosted atop public cloud providers.
VMware VM’s in AWS and Google
Ravello has offered the ability to run unmodified VM’s within AWS or Google Compute infrastructure since the birth of the company’s initial service. The typical use case has been testing and development work. The basic premise is that infrastructure engineers or developers could upload their unmodified VMware images (VMDK) to Ravello and then use a Visio-like tool to draw connectivity between the systems. Based on users cost and performance preferences Ravello would then run the virtual machines on either Google or Amazon cloud services.
The service is made possible by Ravello’s HVX hypervisor. The HVX hypervisor is an overlay that performs binary level virtualization between the VMware VMDK standard and the underlying Cloud service. The level of abstraction is remarkable as both Google and Amazon use different underlying hypervisors to provide VM’s for their clouds.
The engineer feat
The original service’s value proposition is to run VMDK’s on public clouds. Ravello has engineered the ability to
virtualize emulate AMD-V and Intel VT hardware virtualization support. As a result, Ravello runs an additional level of virtualization. Ravello can now run hypervisors within public cloud providers. During a press demo, Ravello had a who’s who of VMware subject matter experts put the service through the paces. Below is a sampling of the most impressive labs.
- William Lam created a 64-node VSAN cluster
- Scott Lowe ran a lab with 250 vSphere hosts
- Mike Preston ran lab in which he vMotion’d between Amazon and Google
- Alastair Cooke created an Autolab vSphere 6 (to be released to the community for use outside of Ravello)
- Simon Gallagher ran vCenter, vRA, vRO, Horizon, ITBM, SRM, etc.
So, the question becomes what’s next? There is a slew of use cases that run through my mind. There are some obvious ones such as testing configuration changes and upgrades to vSphere environments. Another is proof of concepts for not only VMware but mixed environments. A true test of the abstraction of the virtualization support will be the ability to run any Type-1 hypervisors.
Ravello is getting closer to the dream of abstracting the data center. I think HVX has uses outside of nested labs. As the technology matures, I can see a market for the raw HVX technology. Enterprises needing to rapidly deploy data centers can use HVX as the base and apply their abstracted design upon any bare metal underlay.
Hyperconverged use case
I could also see a company such as Nutanix either licensing or buying Ravello’s technology. Much has been made of Nutanix’ desire to build their hypervisor. I question the value to a customer for yet another hypervisors. However, if Nutanix built a hyperconverged platform that could run any VM in its native format atop of their hardware would be a compelling usage.