wasm-dotnet: An HCL repository from sebagomez

WASM-DOTNET

Personal playground for wasm and .net

After seeing this talk from Steve Sanderson I thought I should give it a try and start playing around with WASM.

We'll see where this takes me

After a huge amount of time and many trials and errors this is my current setup.

I have a local minikube ckuster in WSL2 with two nodes, one allows X64 apps and the other is only for WASM (via kwasm). Both nodes are tainted.

I also have a simple console app that will try to find the nth fibonacci number and will at the end print the number and how long it took to find it.

	  static long Fibonacci(int n)
    {
        if (n <= 0)
        {
            return 0;
        }
        else if (n == 1)
        {
            return 1;
        }
        else
        {
            return Fibonacci(n - 1) + Fibonacci(n - 2);
        }
    }

I read somewhere (can't find the link) that this is a good example of a processor consumming algorithm to test performance. So in this case, this vary same code will be compiled for dotnet (x64) and wasm (using wasmtime)

I installed wasmtime in WSL2 using the following command

curl https://wasmtime.dev/install.sh -sSf | bash

To build I also had to install the wasi sdk and export this envionment variable

export WASI_SDK_PATH=/home/seba/.wasi-sdk/wasi-sdk-20.0

I then created a couple of docker files to have two images with the same app, again, one for dotnet and the other one for wasi. Those images are pushed to my personal repo at Docker Hub.

I finally created a couple of Kubernetes jobs that will pull execute the app with the same argument, in this case 40.

After applying those Jobs you can see that wou'll have two pods

k get pod
NAME                 READY   STATUS      RESTARTS   AGE
hello-dotnet-chvjk   0/1     Completed   0          15m
hello-wasm-zchcg     0/1     Completed   0          15m

Once both completed you can check the logs of each to compare the diference in execution time.

❯ k logs hello-dotnet-chvjk 
It took 00:00:00:674 to calculate the 40th Fibonacci number (102334155) in X64!

❯ k logs hello-wasm-zchcg 
It took 00:00:13:757 to calculate the 40th Fibonacci number (102334155) in Wasm!

So, you can see, x64 took less than a second while WASM took over 13 seconds to find the same number :(

What does that mean? Not much really, I don't have a good environment with a fully WASI native node, so the next step is trying to reproduce this in an AKS clsuter to see if I get better results. I would expect, because of all the buzz about it, that wasm beats dotnet in time; but we'll see.

I'll follow this for seeting up the cluster first with Terraform https://learn.microsoft.com/en-us/azure/aks/learn/quick-kubernetes-deploy-terraform?tabs=bash&pivots=development-environment-azure-cli

After running the example from that link I have my local kubectl talking to a EKS cluster with 3 nodes. (I then updated the default node_count value to 1 and applied again)

I'll then follow this example to add a WASM node to the cluster: https://learn.microsoft.com/en-us/azure/aks/use-wasi-node-pools

Get the resource and cluster name with the following command

resource_group_name=$(terraform output -raw resource_group_name)
cluster_name=$(terraform output -raw kubernetes_cluster_name)

And then add the WASI node pool

az aks nodepool add --resource-group $resource_group_name --cluster-name $cluster_name --name mywasipool --node-count 1 --workload-runtime WasmWasi

Label the nodes

k label node aks-agentpool-18894116-vmss000000 run=dotnet
k label node aks-mywasipool-45235557-vmss000000 run=wasm

sebagomez/wasm-dotnet

WASM-DOTNET