bpftime: userspace eBPF runtime
ToolsComments
The GPU kernel support is so intriguing... but wouldn't the latency of moving data between the GPU and a userspace runtime kill the speed gains?
Why are we talking about portability when this is actually about the death of the monolithic kernel in cloud native? Who wants to manage kernel permissions in a scaled Kubernetes cluster?
What is the projected performance hit for the runtime? Security gains are irrelevant if the latency makes the observability tool unusable.
If the primary goal is security in restricted environments, wouldn't it be more effective to use a minimal hypervisor? A userspace runtime might introduce a new attack surface that a thin kernel wrapper avoids.
This is actually useful for those of us on Windows who spend half our day fighting with WSL2 just to get basic tooling working. A runtime that doesn't require a full VM boot is a huge win.
This approach is similar to how WebAssembly provides a portable execution environment. It shifts the focus from hardware-specific hooks to a standardized bytecode format.
The included verifier is a nice touch. It could let developers iterate on their logic in a safe sandbox without risking a kernel panic on their local machine.