Hacker News Reader

Hacker News

Comment by frik | Hacker News Reader

Parent Full thread

frik

"There are multiple competing factors when it comes to choosing the right number of processes.

- For security, more processes is better; one per tab is probably ideal.

- For crash protection, ditto.

- For responsiveness, one process per CPU core is probably best.

- For memory usage, one process is probably best."

as a dev wrote in the mailing list: https://groups.google.com/forum/#!topic/mozilla.dev.platform...

For various technical reasons Firefox e10 will support a few sub-processes in near future, but nothing is set in stone. Rust based Servo supports (afaik) dozens of processes, as we know it from Chrome and IE.

View on HN

With crashes, though rare in Chrome, when one does happen it usually kills all my Chrome windows instead of a single tab as is touted :/

It depends which process crashes. Usually and rare a sub-process crash which means a few tabs need a page reload. If your main process crash (which should be very very rare) then I would suspect you have a faulty memory module in your device or you have little free memory or your operating system has not optimal scheduler (e.g. server OS on a desktop).

I've seen it irregularly on multiple machines, Windows 7, Windows 8 and now Mac OS X. It's actually really soured me on the whole idea of a multi-process architecture. I understand the theoretical benefits, but I don't see messages that a particular tab crashed that much more often than that the entire browser crashed.

"Little free memory" means that Chrome has eaten all my ram though.

Chrome is really lagging behind Firefox in this, as it has no backgrounding / unloading process that I'm aware of to maintain a reasonable memory footprint.

It also attempts to render every tab at once when restoring after a crash - another obvious problem.

Let me guess, it's a process like Flash that kills all of your tabs?

That's why having proprietary technology in browsers sucks because you can't really implement it the way you want to (as a browser vendor).

If you can reproduce that, I'm sure the Chrome team would be very, very interested in that.

Or maybe not, I think they have automated crash reporting (I even used their library, breakpad) but my Chrome has been very consistent in crashing on Windows wakeup for over a year now.

This summer I'm going to be working on an implementation of software based fault isolation for my masters degree. We are looking to sandbox a web server without the need to partition via processes and incur expensive interprocess communication and context switching. I don't see why it couldn't work for Firefox either.

Here is the original paper: https://crypto.stanford.edu/cs155/papers/sfi.pdf

Reply on news.ycombinator.com