Hacker News Reader

Comment by tmysl | Hacker News Reader

This is an awesome PoC for expanding on to ARM CPUs but i don't see the Pi being an efficient virtualization platform. Once you introduce the CPU/memory overhead of 2-3 vms will quickly degrade any meaningful CPU performance.

View on HN

I'm not sure what use case VMWare is aiming at, but Xen is on loads of embedded systems now. These aren't large server VMs; typically they'd be some system where they have (say) one VM as a realtime OS controlling a piece of hardware, and another providing the UI. Under such a system you'd typically do a hard partition, assigning a fixed number of vcpus and memory to each VM.

The LF Edge project is specifically targeted at something about the size of an RPI, and wants to use virtualization for that case:

https://www.lfedge.org/

I guess VMWare is aiming at server market. Here is the summary section of OP:

> Over the past several years, you've seen us demonstrate our virtualization technology on the Arm platform across several use cases, everything from running mission critical workloads on a windmill, to running on the SmartNIC, to running on AWS Graviton in the cloud.

Right, but would you ever run VMWare on a fleet of RPis as a cloud of some sort? I mean, I guess you could, but it seems like a rather strange combination.

You can roll out an virtualisation solution on raspis to many many many devices, perfect for making it rock solid. The arm server market is still very small, you will not haavy many customers able to test it, and much less risking using an still unstable solution.

This is a perfect entry to the arm server market.

A cluster of ESXi servers doing vSAN and HA/FT is just another technique for making a given application highly reliable. There might be "better" ways to do it, but I look at this and see a very straightforward way to turn half a dozen Raspberry Pis into five nines of uptime (or more) for an embedded device.

Actually I misread this. VMWare is specifically saying that they've run ESXi-Arm on windmills. So it's exactly the same thing: They seem to be thinking about going embedded, and RPi turns out to be a good board for a lot of industrial use cases.

The Pi is a good testbed for validating the technology, but I expect this to be more focused on much, much beefier boards meant for datacenters. May as well start getting the bugs out now.

I’ve entirely moved away from VMs at home to Docker. Their ecosystem is quiet robust (maybe not the arm one) and even if there isn’t a docker image for what I want I can usually use the Alpine or Arch one as a base and with a few lines of config get something stood up.

Im on board with the docker model here, a proper kubernetes deployment with shared storage will afford you the same fault tolerance vsan/vmotion would

As a proof-of-concept the RPi is really easy to get your hands on. The same iso they distribute will run on some of the more niche 16 core A72 boards, but those are a fair bit more tailored to the deployment ( SFP+ NICs etc, ) and more expensive

Might as well get the software ready now for the RPi 5 and beyond.

Reply on news.ycombinator.com