The Linux kernel has all the tools needed to manage CPU, RAM, disk, and network resources—but it manages almost nothing related to the GPU, which is the most expensive component in an AI server. Filling that gap are: cgroup dmem, sched_ext, DRA, GPU checkpoints, CXL tiering, microVMs for agents,…
The Linux kernel has all the tools needed to manage CPUs, RAM, disk, and networking—but it does almost nothing to manage GPUs, the most expensive component in an AI server. Filling that gap are: cgroup dmem, sched_ext, DRA, GPU checkpoints, CXL tiering, microVMs for agents, and confidential…
The Linux kernel has all the tools needed to manage CPU, RAM, disk, and network resources—but it does almost nothing to manage the GPU, the most expensive component in an AI server. Filling that gap are: cgroup dmem, sched_ext, DRA, GPU checkpoints, CXL tiering, microVMs for agents, and…
Apple Silicon’s success isn’t due to a single groundbreaking invention, but rather to its control over the entire technology stack—chips, operating systems, compilers, and foundries. The mechanisms behind this, and lessons for other companies.
The scheduler divides the CPU into very small time slices. Understanding this mechanism explains why the computer runs slowly.
Each process has its own contiguous address space. This virtualization layer is the foundation of security and stability.
Computers are a hundred times faster than they were twenty years ago, but you still have to wait for apps to open. The problem isn't the hardware.
A container is simply a process that is often isolated from the rest of the system. Understanding this means understanding its security limitations.
System directories are read-only; updates are applied as a whole, and if they fail, the system reverts to the previous version with a single reboot. The trade-off is convenience.
Load a small program that runs directly in the kernel—it’s safe and requires no reboot. This is the biggest change in Linux over the past decade.
Instead of calling the system for each operation, the program and the kernel communicate through two shared buffers. The number of context switches is significantly reduced.
It’s not just a new protocol. Wayland completely does away with the thirty-year-old intermediary server model, along with the way applications share the screen.
