3 ms·
Sometimes I feel we went about embracing virtualization the wrong way. We added instructions and changed silicon to efficiently run multiple, isolated OSes on a
by jeswin 15y ago
Sometimes I feel we went about embracing virtualization the wrong way. We added instructions and changed silicon to efficiently run multiple, isolated OSes on a single computer.
The bigger problem that needed fixing was how to make the hardware scale transparently. I don't have any solutions to this of course. But I do imagine that someday, a good team might build transparent virtualization that scales over a network of connected nodes without having to code this explicitly.
- rwmj 15y agoIn some senses we have this already: with some clouds you just have to boot a node from a tiny PXE image or a USB key, and it'll discover the cloud, join in, and VMs will migrate onto it. Unfortunately physics simply won't permit transparent scaling of workloads across nodes. For any application involving communication between parts (anything not "embarrassingly parallel") messaging between the parts gets slower and harder as the parts get further away and there are more of them communicating. 1000 nodes have a million possible connections between them, so you can't directly wire them. You need a switching fabric, so your messages have to deal with multiple hops and contention -- slower and harder.