In my post last year, Moving to SUSE I said I would spend more time explaining what Hyperconverged Infrastructure (HCI) was.
The way I would describe HCI is that it’s a way of structuring a data center so that the storage, compute, and networking resources are all bundled into the same boxes, and then software abstracts over it, allowing nodes to use resources of other machines.
An example of something that’s not HCI is a SAN, or a Storage-Area Network.
I want to share something that I’ve been working on for the last three months. Release 1.1.0-rc3 now has working PCI passthrough. Any PCI device in your cluster can now be passed through the hypervisor directly to a VM. This allows virtual machines to directly control a device, like a GPU, without any expensive virtualization layer in the way.
In the diagram above, the guest (VM) has direct memory access to the host’s memory for the device.
After 3.6 years at Amazon, I’ve decided to move to another company to work on open source software. The company is SUSE, they make a Linux distribution and sell cloud computing services, among other things.
At Amazon I worked on AWS Elemental’s Live (video streaming software), a small video encoding device, and the Amazon Scout robotics project. At Scout I built software that controls the robots when autonomy fails, and I worked on safety features and performance optimization of a distributed application that ran on a cluster of servers managed by AWS Fargate.