The thing is it's easy to define free, unused memory. But a lot of the used memory is your system caching stuff that would be free if you needed more than what's actually free. So you can see you have 1g of free memory out of your 4g, but then you allocate 3g and it will do without a sweat and you'd be confused. So you have to go and dig for what those caches are and report that they're effectively free too.
> The “cached” memory includes tmpfs and ramfs for seemingly no reason.
If you're curious why that is by the way, it's because that's actually how these are implemented (tmpfs/ramfs is just a mount to a filesystem where the files never get marked clean[1])
AFAIK the only way for you to figure out how much of your disks is actually cached involves enumerating all tmpfs and ramfs mounts, summing their sizes, and subtracting the sum from the cache size reported by the kernel.
Well, what's the alternative? Invent a new type of memory reservations specifically to account for tmpfs/ramfs mounts? That'd violate your own stated desired goal of
> The OS should just expose a counter for available memory instead of having applications understand every type of memory reservation.
I think a memory pressure indicator is useful. For example, if you’re writing a garbage collector, you can choose to hold off from returning pages back to the kernel when the system is under low pressure, and do this more often under high pressure.
That's very, very marginally useful. If your application recently ran GC when the system was under low pressure and freed a bunch of pages (and still holds to them), and then some other application suddenly acquires a lot of memory, your application won't free those pages until next GC happens. Which most likely won't until you've used up all those freed pages, and at that point a lot of swapping will already have had happen.
Explicitly set per-application upper bounds à la GOMEMLIMIT and Java's -Xmx, together with cgroup's enforcement, seem to be much more useful in practice.
Or there was a proposal for some sort of SIGLOWEM signal that would be sent to all processes in the system (with default disposition of "do nothing") that'd allow applications to release some of its non-vital memory holdings: also a much more timelier notification.
That metric would give you a number of bytes which can be used for pages not backed by files, but it won't give you actual memory usage statistics:
It won't count executable pages and memory-mapped file use as "used" memory, so your system might display gigabytes "free" when it's starving, executables getting paused when code pages are paged-in from disk.
It's just less useful than what's displayed now. "Everyone is doing it wrong" is usually a signal that you're missing something.
The kernel keeps track of active/inactive pages by scanning page tables during the reclaim process. It only scans until it finds enough inactive pages to reclaim. Scanning all pages all the time is very expensive from a performance point of view.
