Hello Ondrej...
Thanks for taking the time.
On 27/04/16 17:13, Ondrej Kozina wrote:
> I observe that putting swap on top of dm-crypt (tested in current 4.6-rc
> kernel) have some unexpected side effects. Most notably it seems to me
> that oom killer joins the party quite prematurely even when there
> remains plenty of free swap space (almost all). But I didn't experience
> any permanent system freeze or crash. Just, let us dig deeper for a
> while...
Yes, 4.6-rc* behaves different from 4.4.x and >= 4.5-rc3 but it still
does not work correctly at all. You may want to try the little test
program I created and attached. I used it quite a lot to stress the system.
With that, you should have no problem seeing the described behavior on
the affected kernels. You may have to adjust the sleep a bit, depending
on your machine, though.
If there is anything else I can do to help, please let me know.
Have a nice day,
Matthias
--
Dipl.-Inf. (FH) Matthias Dahl | Software Engineer | binary-island.eu
services: custom software [desktop, mobile, web], server administration
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
int main(int argc, char** argv)
{
unsigned long chunkSize = 1 << 20; // 1 MiB per \ total mem alloc
unsigned long maxRuns = 8 * (1 << 10); // 2^10 runs / of 8 GiB
void* buffer = NULL;
if(argc == 3)
{
chunkSize = atol(argv[1]);
maxRuns = atol(argv[2]);
}
unsigned long maxMemAllocInMB = (chunkSize * maxRuns) / (1 << 20);
unsigned long curMemAllocInBytes = 0;
for(unsigned int i = 0; i < maxRuns; ++i)
{
buffer = malloc(chunkSize);
memset(buffer, 0, chunkSize);
curMemAllocInBytes += chunkSize;
printf("%lu MB / %lu MiB allocated (%lu bytes).\n", (curMemAllocInBytes / (1 << 20)), maxMemAllocInMB, curMemAllocInBytes);
usleep(500);
}
return 0;
}
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
