firecracker VS wasmtime

This architecture leverages microVMs for rapid scaling and high-density workloads. But does it work for GPU? The answer is no. You can look at the old 2019 GitHub issue and the comments to it to get the bigger picture of why it is so.

Hi, I'm the CEO of the company that built this SDK.

We're a company called E2B [0]. We're building and open-source [1] secure environments for running untrusted AI-generated code and AI agents. We call these environments sandboxes and they are built on top of micro VM called Firecracker [2].

You can think of us as giving small cloud computers to LLMs.

We recently created a dedicated SDK for building custom code interpreters in Python or JS/TS. We saw this need after a lot of our users have been adding code execution capabilities to their AI apps with our core SDK [3]. These use cases were often centered around AI data analysis so code interpreter-like behavior made sense

The way our code interpret SDK works is by spawning an E2B sandbox with Jupyter Server. We then communicate with this Jupyter server through Jupyter Kernel messaging protocol [4].

We don't do any wrapping around LLM, any prompting, or any agent-like framework. We leave all of that on users. We're really just a boring code execution layer that sats at the bottom that we're building specifically for the future software that will be building another software. We work with any LLM. Here's how we added code interpreter to Claude [5].

Our long-term plan is to build an automated AWS for AI apps and agents.

Happy to answer any questions and hear feedback!

[0] https://e2b.dev/

[1] https://github.com/e2b-dev

[2] https://github.com/firecracker-microvm/firecracker

[3] https://e2b.dev/docs

[4] https://jupyter-client.readthedocs.io/en/latest/messaging.ht...

[5] https://github.com/e2b-dev/e2b-cookbook/blob/main/examples/c...

As far as I know, Fly uses Firecracker for their VMs. I've been following Firecracker for a while now (even using it in a project), and they don't support GPUs out of the box (and have no plan to support it [1]).

I'm curious to know how Fly figured their own GPU support with Firecracker. In the past they had some very detailed technical posts on how they achieved certain things, so I'm hoping we'll see one on their GPU support in the future!

[1]: https://github.com/firecracker-microvm/firecracker/issues/11...

I pass through a GPU and USB hub to a VM running on a machine in the garage. An optical video cable and network compatible USB extender brings the interface to a different room making it my primary “desktop” computer (and an outdated laptop as a backup device). Doesn’t get more silent and cool than this. Another VM on the garage machine gets a bunch of hard drives passed through to it.

That said, hardware passthrough/VFIO is likely out of the current realistic scope for this project. VM boot times can be optimized if you never look for hardware to initialize in the first place. Though they are still likely initializing a network interface of some sort.

“MicroVM” seems to be a term used when as much as possible is stripped from a VM, such as with https://github.com/firecracker-microvm/firecracker

According to their own FAQ it is indeed: https://github.com/firecracker-microvm/firecracker/blob/main...

What about microVMs like firecracker?

Dynamic memory management - Firecracker's RAM footprint starts low, but once a workload inside allocates RAM, Firecracker will never return it to the host system. After running several workloads inside, you end up with an idling VM that consumes 32 GB of RAM on the host, even though it doesn't need any of it.

Firecracker has a balloon device you can inflate (ie: acquire as much memory inside the VM as possible) and then deflate... returning the memory to the host.

https://github.com/firecracker-microvm/firecracker/blob/main...

Very few things in those companies are being written in Rust, and half of those projects chose Rust around ideological reasons rather than technical, with plenty of 'unsafe' thrown in for performance reasons

https://github.com/firecracker-microvm/firecracker/search?q=...

The fact that 'unsafe' even exists in Rust means it's no better than C with some macros.

Don't get me wrong, Rust has it's place, like all the other languages that came about for various reasons, but it's not going to gain wide adoption.

Future of programming consists of 2 languages - something like C that has a small instruction set for adopting to new hardware, and something that is very high level, higher than Python with LLM in the background. Everything in the middle is fodder.

https://github.com/firecracker-microvm/firecracker is the one that comes to mind, but most of these are internal.

Just a documentation change, fortunately:

https://github.com/bytecodealliance/wasmtime/commits?author=...

They've submitted little documentation tweaks to other projects, too, for example:

https://learn.microsoft.com/en-us/cpp/overview/whats-new-cpp...

I don't know whether this is a formerly-legitimate open source contributor who went rogue, or a deep-cover persona spreading innocuous-looking documentation changes around to other projects as a smokescreen.

WebAssembly is extremely portable. WebAssembly runs on: all major web browsers, V8 runtimes like Node.js, and independent Wasm runtimes like Wasmtime, Lucet, and Wasmer.

cpu: 4 disk: 60 memory: 12 arch: host hostname: colima autoActivate: true forwardAgent: false # I only tested this with 'docker', not 'containerd': runtime: docker kubernetes: enabled: false version: v1.24.3+k3s1 k3sArgs: [] network: address: true dns: [] dnsHosts: host.docker.internal: host.lima.internal # Added: # - containerd-snapshotter: true (meaning containerd will be used for pulling images) docker: features: buildkit: true containerd-snapshotter: true vmType: vz rosetta: true mountType: virtiofs mountInotify: false cpuType: host # This provisioning script installs build dependencies, WasmEdge and builds the WASM runtime shims for containerd. # NOTE: this takes a LOOONG time! provision: - mode: system script: | [ -f /etc/docker/daemon.json ] && echo "Already provisioned!" && exit 0 echo "Installing system updates:" apt-get update -y apt-get upgrade -y echo "Installing WasmEdge and runwasi build dependencies:" # NOTE: packages curl, git and python3 already installed: apt-get install -y make gcc build-essential pkgconf libtool libsystemd-dev libprotobuf-c-dev libcap-dev libseccomp-dev libyajl-dev libgcrypt20-dev go-md2man autoconf automake criu pkg-config libdbus-glib-1-dev libelf-dev libclang-dev libzstd-dev protobuf-compiler apt-get clean -y - mode: user script: | [ -f /etc/docker/daemon.json ] && echo "Already provisioned!" && exit 0 # # Setting vars for this script: # # Which WASM runtimes to install (wasmedge, wasmtime and wasmer are supported): WASM_RUNTIMES="wasmedge wasmtime wasmer" # # Location of the containerd config file: CONTAINERD_CONFIG="/etc/containerd/config.toml" # # Target location for the WASM runtimes and containerd shims ($TARGET/bin and $TARGET/lib): TARGET="/usr/local" # # Install rustup: # echo "Installing rustup for building runwasi:" curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- --default-toolchain none -y source "$HOME/.cargo/env" # # Install selected WASM runtimes and containerd shims: # [[ -z "${WASM_RUNTIMES// /}" ]] && echo "No WASM runtimes selected - exiting!" && exit 0 git clone https://github.com/containerd/runwasi echo "Installing WASM runtimes and building containerd shims: ${WASM_RUNTIMES}:" sudo mkdir -p /etc/containerd/ containerd config default | sudo tee $CONTAINERD_CONFIG >/dev/null for runtimeName in $WASM_RUNTIMES; do case $runtimeName in wasmedge) echo "Installing WasmEdge:" curl -sSfL https://raw.githubusercontent.com/WasmEdge/WasmEdge/master/utils/install.sh | sudo bash -s -- -p $TARGET echo echo "`wasmedge -v` installed!" ;; wasmtime) echo "Installing wasmtime:" curl -sSfL https://wasmtime.dev/install.sh | bash sudo cp .wasmtime/bin/* ${TARGET}/bin/ rm -rf .wasmtime echo "`wasmtime -V` installed!" ;; wasmer) echo "Installing wasmer:" curl -sSfL https://get.wasmer.io | sh sudo cp .wasmer/bin/* ${TARGET}/bin/ sudo cp .wasmer/lib/* ${TARGET}/lib/ rm -rf .wasmer echo "`wasmer -V` installed!" ;; *) echo "ERROR: WASM runtime $runtimeName is not supported!" exit 1 ;; esac cd runwasi echo "Building containerd-shim-${runtimeName}:" cargo build -p containerd-shim-${runtimeName} --release echo "Installing containerd-shim-${runtimeName}-v1:" sudo install ./target/release/containerd-shim-${runtimeName}-v1 ${TARGET}/bin sudo ln -sf ${TARGET}/bin/containerd-shim-${runtimeName}-v1 ${TARGET}/bin/containerd-shim-${runtimeName}d-v1 sudo ln -sf ${TARGET}/bin/containerd-shim-${runtimeName}-v1 ${TARGET}/bin/containerd-${runtimeName}d echo "containerd-shim-${runtimeName} installed." cd .. echo "[plugins.\"io.containerd.grpc.v1.cri\".containerd.runtimes.${runtimeName}]" | sudo tee -a $CONTAINERD_CONFIG >/dev/null echo " runtime_type = \"io.containerd.${runtimeName}.v1\"" | sudo tee -a $CONTAINERD_CONFIG >/dev/null done echo "containerd WASM runtimes and shims installed." # # Restart the systemctl services to pick up the installed shims. # NOTE: We need to 'stop' docker because at this point the actual daemon.json config is not yet provisioned: # echo "Restarting/reloading docker/containerd services:" sudo systemctl daemon-reload sudo systemctl restart containerd sudo systemctl stop docker sshConfig: true mounts: [] env: {}

When you say wasm container, you mean something like wasmtime that provides a non-browser wasm runtime?

https://github.com/bytecodealliance/wasmtime

We can't run this directly from the command line unless we install some runtime like wasmtime:

The roadmap I linked above. The WASI folks have done a poor job at communicating, no doubt, but I'm surprised someone like yourself literally building a competitor spec isn't following what they are doing closely.

Just for you I did some googling: see here[0] for the current status of WASI threads overall, or here[1] and here[2] for what they are up to with WASI in general. In this PR[3] you can see they enabled threads (atomic instructions and shared memory, not thread creation) by default in wasmtime. And in this[4] repository you can see they are actively developing the thread creation API and have it as their #1 priority.

If folks want to use WASIX as a quick and dirty hack to compile existing programs, then by all means, have at it! I can see that being a technical win. Just know that your WASIX program isn't going to run natively in wasmtime (arguably the best WASM runtime today), nor will it run in browsers, because they're not going to expose WASIX - they're going to go with the standards instead. so far you're the only person I've met that thinks exposing POSIX fork() to WASM is a good idea, seemingly because it just lets you build existing apps 'without modification'.

Comical you accuse me of being polarizing, while pushing for your world with two competing WASI standards, two competing thread creation APIs, and a split WASM ecosystem overall.

[0] https://github.com/bytecodealliance/jco/issues/247#issuecomm...

[1] https://bytecodealliance.org/articles/wasmtime-and-cranelift...

[2] https://bytecodealliance.org/articles/webassembly-the-update...

[3] https://github.com/bytecodealliance/wasmtime/pull/7285

[4] https://github.com/WebAssembly/shared-everything-threads

Thanks for the question!

Spin could definitely run in more places than what we have pre-built binaries for. Specifically, we could run on all platforms Wasmtime supports today (https://github.com/bytecodealliance/wasmtime/releases/tag/v1...), including RISC and S390X, for example.

And while we have been experimenting a bit with running Spin on RISC, we haven't really had the bandwidth or requirement to build a production build for those yet.

Are you interested in a specific operating system or CPU architecture? Would love to understand your scenario.

> I used to think we'd eventually get to capability based security, but now I see we'll always be stuck with application permission flags, the almost worthless bastard cousin, instead.

My hope is that WASI will introduce capability based security to the mainstream on non-mobile computers [0] - it might just take some time for them to get it right. (And hopefully no half-baked status-quo-reinforcing regressive single—runtime-backed alternatives win in the meantime.)

[0]: https://github.com/bytecodealliance/wasmtime/blob/main/docs/...

WasmTime finished finished the RFC for the implementation details in June: https://github.com/bytecodealliance/wasmtime/issues/5032

[3]: https://github.com/bytecodealliance/wasmtime/pull/2611