ModalAyush RanjanAnnouncement
Modal becomes a gVisor maintainer
Summary
Google is donating gVisor, including its name and trademarks, to the CNCF, moving the project to a maintainer-led governance model. Modal, alongside Ant Group and Tines, has committed to join gVisor’s maintainers, with OpenAI, Tencent, and NVIDIA continuing their contributions. Modal launches over 40 million gVisor containers daily for AI workloads, including GPU inference, multi-node training, and coding agents. gVisor adds a separate application kernel called the Sentry between the container and the host, implementing the Linux system call interface itself for stronger isolation without hardware virtualization. Modal has upstreamed patches for GPU support via nvproxy, RDMA via rdmaproxy, filesystem performance improvements, and checkpoint/restore capabilities.
Context
Modal runs over 40 million gVisor containers daily for AI workloads including GPU inference, multi-node training, and coding agents. gVisor provides a security boundary between tenants on shared machines without requiring hardware virtualization, making it suitable for cloud VMs and bare-metal GPU servers.
Approach / What changed
Modal contributes upstream patches to gVisor for GPU support (nvproxy), RDMA (rdmaproxy), filesystem performance, and checkpoint/restore, and joins the project’s maintainers alongside Ant Group and Tines after Google donates gVisor to the CNCF.
Takeaways
gVisor provides a security boundary between tenants on shared machines without requiring hardware virtualization, making it suitable for cloud VMs and bare-metal GPU servers.
Modal’s upstream contributions include nvproxy for GPU support, rdmaproxy for RDMA, and filesystem optimizations that improved SQLite performance by 2.7x and PyTorch DataLoader throughput from 2.9 to 22 GB/s.
gVisor can boot a sandbox in about 25 ms with only 12 MB of memory overhead, and supports checkpoint/restore of entire sandboxes including GPU memory.
FAQ
How does gVisor isolate containers from the host kernel?
gVisor adds a separate application kernel called the Sentry, written in Go, between the container and the host. The Sentry implements the Linux system call interface itself and asks the host for only a small, filtered set of operations, so the app never talks to the host kernel directly.
How does gVisor support GPU access in sandboxes?
gVisor's nvproxy lets a sandboxed process use NVIDIA GPUs through a proxy that understands and validates the driver's ioctl interface, instead of exposing the raw driver to the container. GPU memory and compute stay native.
What is gVisor's checkpoint/restore capability?
gVisor can checkpoint an entire sandbox (every process, file, socket, and GPU memory) and restore it on another machine. Modal uses this for memory snapshots, restoring about 670,000 gVisor containers from snapshots each day.
Source and freshness
- Source
- modal.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.