New: High paging activity as soon as the swap is touched (with steps and code to reproduce it)

https://bugzilla.kernel.org/show_bug.cgi?id=200105

Bug ID: 200105
Summary: High paging activity as soon as the swap is touched
(with steps and code to reproduce it)
Product: Memory Management
Version: 2.5
Kernel Version: 4.14 4.15 4.16 4.17
Hardware: All
OS: Linux
Tree: Mainline
Status: NEW
Severity: normal
Priority: P1
Component: Page Allocator
Assignee: [email protected]
Reporter: [email protected]
Regression: No

Hi.

Under any desktop environment, as soon as the swap is touched the kernel starts paging out gigabytes of memory for no apparent reason causing the system to freeze for at least a minute.

This happens with normal desktop usage but I'm able to reproduce this behavior systematically by using the code included at the end of this report, let's call it "memeater". memeater takes three parameters: totalmem, chunksize, sleeptime. It allocates (and initializes to 0) totalmem bytes of memory in chunks of chunksize bytes each. After each chunk allocation it sleeps for sleeptime seconds. If sleeptime=0, it does not sleep in between chunk allocations. After totalmem bytes of memory have been allocated it sleeps indefinitely.

To reproduce the behaviour using memeater:

1) start a desktop environment (in my case KDE plasma).
2) invoke "memeater x y 0" to bring the system to the brink of swapping. I usually execute "memeater x y 0" multiple times where x is 4 gigs and y is 20 megs.(I've got 16GB of memory so I usually go up to 14/15 GB of used memory).
3) invoke "memeater x y 1" one last time so that it slowly fills up the memory with small chunks of a few megs each (again, I tried it with x=4 gigs and y=20 megs).

When the last memeater fills the memory (and keeps allocating chunks) I would expect the mm system to swap out few megabytes worth of pages to accomodate the new requests slowly coming from the last memeater.
However, what actually happens is that the mm starts to swap out gigabytes worth of pages completely freezing the system for one or two minutes.

I've verified this on various desktop systems, all using SSDs. Obviously, I'm willing to provide more info and to test patches.

####### memeater #######################à

#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>

int main(int argc, char** argv) {
    long int max = -1;
    int mb = 0;
    long int size = 0;
    long int total = 0;
    int sleep_time = 0;
    char* buffer;

    if(argc > 1)
      {
        max = atol(argv[1]);
        size = atol(argv[2]);
        sleep_time = atoi(argv[3]);
      }
    printf("Max: %lu bytes\n", max);
    while((buffer=malloc(size)) != NULL && total < max) {
        memset(buffer, 0, size);
        mb++;
        total=mb*size;
        printf("Allocated %lu bytes\n", total);
        if (sleep_time) sleep(sleep_time);
    }      

sleep(3000000);
return 0;
}

Signature · 2 lines

You are receiving this mail because:
You are the assignee for the bug.

Quoted 2 lines

I've verified this on various desktop systems, all using SSDs. Obviously, I'm willing to provide more info and to test patches.