2 ms·
WSL1 -> WSL2 wasn't only about filesystem performance. It looks like the Fuschia people are going to relearn those lessons the hard way. Namely, that it is much
by roca 5y ago
WSL1 -> WSL2 wasn't only about filesystem performance. It looks like the Fuschia people are going to relearn those lessons the hard way. Namely, that it is much less work to make a real Linux kernel work well as a virtual guest than to reimplement compatible Linux kernel semantics in a host-OS-specific compatibility layer, and the former has better performance too. Also a lot of the work for the former is shared with other vendors whereas for the latter you're on your own.
- mwcampbell 5y agoWhat about Illumos's LX-branded zones? Bryan Cantrill has bragged about how complete that implementation is, about how it could even run something as exotic as the 100-language quine relay [1]. Sure, Joyent (who did most of the recent-ish work on this) ended up exiting the public cloud business, but surely that had more to do with the brutal winner-take-all nature of that space than any failure of the technology. [1]: https://github.com/mame/quine-relay https://github.com/mame/quine-relay
- roca 5y agoThat Quine thing probably uses almost none of the Linux kernel API. Some of the language runtimes do some interesting things, but even fancy ones like V8 are not that challenging in terms of kernel APIs. So I don't know anything about Illumos Linux zones, but if they run rr or docker or Firefox or WINE I'll be stunned. (Note that Firefox, rr and docker (by default) use seccomp-bpf syscall filtering which would be an absolute beast to implement in your Linux compatibility layer.)