Hi. On Fri, Mar 01, 2019 at 02:39:22PM -0300, Bruno Schneider wrote: > On Fri, Mar 1, 2019 at 2:18 PM Reco wrote: > > > > Please post the contents of /proc/meminfo. And "sar -r ALL 1 10", for > > the sake of the completeness. > > $ cat /proc/meminfo > MemTotal: 3975380 kB > MemFree: 1886004 kB > MemAvailable: 2307332 kB
Now that, ladies and gentlemen, is interesting. > Buffers: 49840 kB > Cached: 638404 kB Especially combined with this. > SwapCached: 28688 kB And this. > SwapTotal: 979928 kB > SwapFree: 764888 kB And this. First, there's huge amount of unused (not to be confused with "free") memory on your host. And no, it's not a filesystem's cache (600M), it's really "nothing there"-unused memory which amounts 1880M. Second, you do have *something* in the swap (220M give or take), yet it's 28M of that which got swapped back. Taking it all into the account, I propose the following scenario - not a long time ago there was a process (or a couple of those). Was it the browser, or something written in Java (or Python), or some kind of Virtual Machine - is not relevant, but it did consume RAM. But it got terminated, and what we're seeing here is the aftermath. > There is no sar command on my system. Consider installing it, the package's called sysstat. Comes with useful diagnostic utilities of all shapes and colors, and they *do* have a minimal resource consumption. > > To prevent it: > > > > sysctl -w vm.swappiness=0 > > > > I just read some people arguing that 1 is better than 0 because the > later means "never swap" while the former means "swap if out of RAM". > I'll try 1 to see if it reduces swapping, thanks. Whatever floats your boat. I use "10" personally, but YMMV. Reco
