Hacker News Reader

Comment by greatjack613 | Hacker News Reader

If its not caching to disk, doesn't that mean that ram usage will skyrocket? As it is chrome is a memory hog, imagine now with all of those videos in ram....

View on HN

If the content is ephemeral then it doesn't make sense to keep it on disk, and as long as Chrome doesn't lock the memory range, the OS will swap it to disk anyway if it needs the RAM for something else.

When I watched high-resolution videos on Youtube with Chrome, its process could take about 400 Mb of memory or even more and swap out other processes.

> and as long as Chrome doesn't lock the memory range, the OS will swap it to disk anyway

Swapping in/out is random access and is very slow (< 1 Mb/sec) compared to linear reading/writing unless you have SSD.

Swapping contiguous regions is sequential, and Windows will preemptively write non-locked regions to disk so it can instantly discard them when needed. You can force this behavior by dropping working sets, which has the effect of "converting" used RAM to cached pagefile.

Of course if you're actively using a program, it will swap out other processes, that's the whole idea.

It doesn't make sense to swap out other programs just to load video that will be immediately discarded after displaying. It makes more sense to have limited download and playback buffers and store most of video data on the disk.

When I watched high-resolution videos on Youtube with Chrome, its process could take about 400 Mb of memory or even more and swap out other processes.

I've noticed you've said this twice in this thread like it's a bad thing. This is exactly the behaviour I'd expect and want.

If I'm watching a video I'd want the system to do everything possible to make it as smooth as possible.

I've been running Firefox from tmpfs for at least a decade. (tmpfs has more useful properties than letting that RAM go to Linux disk caches instead, in this case.)

Firefox doesn't need all that much page/object cache, and videos usually seem to be streamed with transient buffering.

If you're playing a video, which is the use case in the article, it wouldn't make sense keep anything in RAM. A minute buffer in either direction seems logical. I am expecting the implementation to be something like that.

Ram is meant to be used. Ram should be 100% utilized

Yes, but if your application assumes that it should use 100% of my RAM, it's being antisocial. Don't assume that I don't have more important software that could be utilizing my RAM in ways more important to me.

That's what drives me up a wall whenever people use the "free RAM is wasted RAM" argument; the assumption that RAM unused by your software is unused by me.

The operating system should be responsible for managing resources like RAM, and applications should be responsible for using the right resources. Caching video is absolutely a RAM thing, since it's expected to go away once the video is done. Caching downloads is a separate matter, since it's expected to stay once finished.

A good litmus test here is what you expect to happen if the machine reboots. For video, I think it's reasonable to lose all video cache, but I think downloads should be resumable. If it shouldn't be there on a reboot, use RAM. If it should, use disk.

There are certainly exceptions, but I think the general rule of thumb is useful.

Android found a kind of solution for this by agressively killing processes (while saving their state). This seems to be effective because developers are not motivated to help competing applications, they want to have all resources for themselves.

That's perhaps great in android, but you can't/shouldn't write code for Linux/Windows/MacOS, where such a system is not in use, and then protest "I'm not the one that sucks." Maybe those OSes should implement such a system, but until they have the "wasted RAM" argument needs to end for anything other than specialty professional software that can reasonably assume itself to be the raison d'être of that computer (e.g. video editing software, CAD, build systems for large software projects, etc.)

Windows has such a system with UWP. I think UbTouch or whatever the Ubuntu phone successor distribution is called has one too.

And I imagine most systems programmers could define an api if they wanted to.

I think this is definitely a case where apps need to be antisocial and operating systems need to peg them in. The age of cooperative multitasking is over. Let the OS decide.

Harming users to coerce OS designers is a pretty shitty plan. Sorry if that's too blunt, but that's simply the way I feel.

I have 4 Gb of RAM and most of time it is utulized over 100%. Not a very pleasant experience with modern Electron apps and reactive HTML5 sites.

Spoken like someone who has truly zero understanding of how RAM usage relates to performance.

Repeat after me: high data locality reduces the amount of cache misses, which drastically reduces memory latency, and is only achievable through systematic memory usage conservativism.

You're assuming this person understands data locality, or memory latency or even memory in general.

Try running Office, Chrome and Teams on a constrained system, then say that again!

You seem to be ignoring a very important factor here, energy consumption. RAM usage has a direct effect on battery usage and battery usage is an important factor and selling point these days.

RAM _capacity_ has a direct effect on energy, usage not nearly much if at all.

Hardware has been around to selectively put banks into low power saving mode, but I've never heard of any commodity OS that supported anything like that. It'd require active organization of physical pages to take advantage of, if common controllers even supported it

Note that most of RAM's power consumption comes from reading and writing (a function of memory clock speed). Allocating the RAM itself is likely to meaningfully increase power draw, but accessing it will.

Are there any benchmarks around showing this? And how much extra power draw it is.

Does it? I was under the impression that all the installed RAM got refreshed all the time whether it holds useful data or not.

The state of bits in RAM has zero effect on energy usage. Caching things in memory allows for lower power consumption, though, by preventing duplication of effort by the CPU.

Edit: Originally said "zero effect on memory usage", which makes very little sense!

Surely not as bad as disk.

I tend to have dozens of tabs open at the same time. However, only a few of them usually have video elements and Chrome could probably discard the media cache for all of them - except for the currently active one.

My assumption here is that the media cache is mainly used for uninterrupted playback in case my network is briefly unavailable (e.g. when on a mobile network).

Reply on news.ycombinator.com