This enables secure and efficient workload orchestration in right-sized microVMs.

Watch @markrussinovich.bsky.social explain it all in "Inside Azure Innovations" (@ 15:45):
ignite.microsoft.com/en-US/sessio... [3/3]

direct access to hardware accelerated devices such as GPUs for full performance.

In Direct Virtualization, the parent orchestrator VM can ask the hypervisor to take its CPU, memory and device resources, and create child VMs with those resources to run untrusted workloads it directly controls. [2/3]

...your code in a child VM on top of a nested hypervisor, which requires maintenance and comes with significant performance penalties.
With DV, you can now run child VMs inside your VM using the Azure hypervisor, with full performance parity and device acceleration. No nested hypervisor required.

WSL runs microVMs, same tech that was initially developed for running hw-virtualized containers (aka Hyper-V “Xenon” containers) on Windows. It exposes a minimal modern (“born in the cloud”) virtual motherboard, plus state mirroring between the two kernels to keep them in-sync. Really cool stuff.

There is a "Skype dial pad" feature in the Teams app (both Windows and iOS), but it is just a link to calling.web.skype.com.

Skype the brand is dead, but Skype the product is still alive; it just merged into Teams. All accounts, contacts, chats, credits etc. are still available in Teams.

I left Amazon because I hated the politics and bureaucracy there, but I was still proud of the work we did. Now, this, just makes me feel sick.

...and now this. Official Trump Store on Amazon. There is no limit to how deep Amazon will kiss his ass if it gives them an advantage.
www.amazon.com/stores/Trump...

So you’ve reached the bargaining stage. I’ve been stuck in depression since January.

XPS isn’t dead. It is just being rebranded as “Max”, starting with the next generation in mid-2025.

That's the beauty of incremental improvements. I enjoy the simplicity, the ROI, the craft, as much as those shiny big projects. We overstate impact of "new". Impact is putting in that extra effort in everyday work, making the right engineering trade-offs, wherever you can find it. 2 MiBs at a time.

On a typical Windows client, savings amount to ~2MiB of nonpaged pool. Not much. When you install the Windows 11 24H2 update, you won't notice it. On scale, for 1.5 billion eligible Windows clients, these 20 lines save 3 PiBs (~2^51) of memory globally. And I've done it 3 times already this year.

Simply growing it would "fix the bug". But at Microsoft's scale, every line of code impacts tens, hundreds of millions of people. One-size-fits-all, doesn't. So, I wrote a heuristic that adjusts the buffer size dynamically based on the OS SKU, processor count, memory size, etc. 20 lines in an hour.

The hypervisor runs everywhere nowadays, from Xbox to Azure, ranging from hw with a few cores and MBs of memory, to giants in Azure with 100s of cores and TBs of memory. It also runs on Windows 10/11, where it's the foundation for virtualization-based security. The buffer needed to grow for Azure.