Current Performance

[victorb]

Looking into deploying npm-on-ipfs together with the rest of the Jenkins infrastructure to make the npm install a bit faster.

However, preliminary testing on my local machine makes me believe that npm-on-ipfs makes the installs slower, even on second installs (where all packages should be cached) and with npm-on-ipfs running on the same machine, compared to just running npm install connected directly to registry.npmjs.com.

[achingbrain]

It might not be faster (since we have to do more work to serve a file) but it should be more reliable. The second install thing is interesting, npm should work out versions from the package-lock.json file and install modules from it's own cache, or did you delete it?

Anyway - I'm happy to look into performance more - do you have any figures for this or a test I can replicate?

[victorb]

The way I tested it was something like this (in js-ipfs & removing package-lock.json as we currently don't have it):

rm -rf node_modules/ package-lock.json || true
npm install --cache=$(mktemp -d) --registry=https://registry.npmjs.com

And repeat again but replacing the registry with local URL.

First run with npm-on-ipfs took 4 minutes, but that's understandable as it has to download the package, add it to IPFS and then serve it. But I would expect the second run to take similar amount of time as normal npm install.

[olizilla]

WIP on tests for performance: #103

Currently only tests for first install speed, but it'd be very useful for on-boarding folks if we can improve that

[andrew]

Some other relevant performance measurements made by @achingbrain over here: zkat/pacote#173 (comment)

The most surprising one was "Local IPFS node that already has the tarball: Avg: 1022.4ms" compared with 830ms to make a http request to a server on the internet for the same content.