Hacker News Reader

Comment by ipsum2 | Hacker News Reader

> Any inactive tabs will be reloaded when you need them.

Anyone have ideas why the tab's memory contents can't be offloaded to SSD and restored instead? Personally I don't like losing state (e.g. filled in forms, scroll position, deleted posts, etc) from using other extensions like Great Suspender.

View on HN

Just a guess but probably something to do with running states in the background but connected to external thingies, hard to offload and restore as-is if the external thingies (or the local one) detect timeouts. Easier to just reload the URL than bother to restore an unusable webpage's state I suppose.

OTOH this is already an issue with you sleep / suspend the machine itself

Don't forget that reloading through the networks gives Google the ability to keep tracking. The requests will go only to that URL, but Google will get it one way or another...

Why would Google need a reload to keep tracking (they already control the browser)? That this feature/implementation was driven by desire for more tracking sounds far fetched to me.

I think it is possible. Typical web process memory size is about 30-500 Mb. Writing 500 Mb sequentially on a spinning HDD takes 5 to 10 seconds, and probably you can discard a lot before saving (like JIT code, JS bytecode, CSS dependency trees, HTTP caches, localstorage etc.), and 500 Mb pages are rare.

For comparison, if you don't unload the page completely, and system gets into swapping then there will be random access pattern with speed about 400 Kb/s on a HDD. So it is faster to unload 500 Mb sequentially than swap out/in 5 Mb.

This is why getting away from optimizing all our software around the disk being the major bottleneck will be a great thing for computing. So many optimizations become possible when your storage isn't in the hot path.

Are you really using a spinny disk? Most SSDs can do 500mb in a second or two.

I considered the worst case, obviously with SSD you can unload pages even faster if you are fine with reducing its lifetime.

If you're developing a widely used web browser you have to take into account the people booting from a hard drive made in 2009.

1. it's a ton of data

2. it's very messy data, it's a bunch of seas of nodes (the DOM, the JS heap, the styling info, etc...)

3. I wouldn't be surprised if it was data which isn't fully isolated (e.g. shared style caches), meaning it first has to be fully forked to resolve into a self-contained bundle

OS have the ability to just freeze and store / restore memory, but I'm not sure they give that capability to use processes.

If I'm not mistaken, the original Word document file format was a RAM dump.

Sure, so were save files for many older games, but it was a dump of the document / game state, which was a fraction of the running program, and probably designed for that in mind in the first place.

Image-based systems (smalltalk, self) would probably be a better model for a live tab, but it still requires things to line up.

As a web app developer? I’ve seen all sorts of ridiculous behaviour that I can’t explain for reasons other than weird wake-from-sleep implementations. I assume that any tab suspension implementations will follow the same patterns.

Reply on news.ycombinator.com