Bulk request fetching about pacote

[boutell]

Bulk stuff as in... big buffer sizes passed to .read()? Just curious what this is about.

(Is it possible to tell a stream to use a bigger buffer size to reduce redundant code?)

[zkat]

@boutell it depends! fs.{read,write}File() are significantly faster than fs.create{Read,Write}Stream(). On the order of ~3x speedup for the former. And that's the case even when you bump things up such that it's a single _write. There's benchmarks over at cacache that test reads and writes both in streaming and bulk mode, too, and cacache itself is already 3-4x faster when doing bulk operations on both read and write. (just clone the repo, do npm i && npm run benchmarks)

The story with network requests is a little different. Probably because of all the stuff npm-registry-client is doing for restarting and parsing bulk requests, streaming: true is ~2x+ faster (more? I had a benchmark somewhere).

So pacote would need to make judicious use of both streams and bulk writes, depending on the data size (which fortunately we usually have, because of Content-Length).

I already have a benchmark suite branch that I need to nudge a little more before it's ready to merge, but that's gonna be super important in finding the right combination for all this :)

[zkat]

Maybe I'll just push it up cause it might start being useful to folks who are interested in doing this sort of optimization golf 😁

[zkat]

#61 There we go. This should help.