Hi ,
This issue can be replicated by executing the above mentioned code on
debian9 and debian10 OS. The compiler arguments I used is g++ debian.cc
(debian.cc is the file name). After compiling we need to check the output
file(i.e Result). In the output file focus on the 'used' field of "free -m"
cmd. You can see the memory usage difference between debian9 and debian10.
To measure memory usage I use the "free -m" command and I focused on the
"used" field to check how much memory has been consumed.
Below is the output of the code on deb10
total used free shared buff/cache
available
Mem: 7914 1789 4306 0 1819 5826
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1789 4306 0 1819 5826
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1791 4304 0 1819 5824
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1791 4304 0 1819 5824
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1793 4302 0 1819 5822
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1793 4302 0 1819 5822
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 1795 4300 0 1819 5820
Swap: 0 0 0
.
.
.
Mem: 7914 4592 1502 0 1819 3022
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 4594 1500 0 1819 3020
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 4594 1501 0 1819 3021
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 4596 1499 0 1819 3019
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 4596 1499 0 1819 3019
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7914 4598 1497 0 1819 3017
Swap: 0 0 0
If you check the memory usage on "used" it is consuming ~4.5G by the end
of the iteration.
Below is the output of the code on deb9.
total used free shared buff/cache
available
Mem: 7923 1843 4128 4 1951 5776
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1843 4128 4 1951 5776
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1843 4128 4 1951 5776
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1843 4128 4 1951 5776
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1843 4128 4 1951 5776
Swap: 0 0 0
.
.
.
total used free shared buff/cache
available
Mem: 7923 1870 4101 4 1952 5750
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1870 4101 4 1952 5750
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1870 4100 4 1952 5750
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1870 4100 4 1952 5750
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1870 4101 4 1952 5750
Swap: 0 0 0
total used free shared buff/cache
available
Mem: 7923 1870 4100 4 1952 5750
Swap: 0 0 0
By the end of the iteration it is consuming ~1.8G
As we see from the output there is a huge memory consumption difference
between deb9 and deb10. I would like to know why this is different, when
the same code runs on different Debian machines (i.e. deb9 and deb10).
On Fri, 1 Jul 2022 at 18:27, Anu Anu <[email protected]> wrote:
> Hi,
> In the below c++ program I am trying to divide a bigger memory block into
> smaller subblock by moving pointers by 1 page. While creating a subblock I
> am
> initializing the variable of a structure which is of 64 bits. I tried to
> run this code on debian10 and debian9. What I observed is in deb9 only 64
> bits memory is consumed but deb10 consumes the entire page which is off 4K.
> Can someone help me understand what causes this difference? Can this be
> related to the way the kernel handles the memory? Thanks for your help in
> advance.
>
> #include <vector>
> #include <stdlib.h>
> #include <fcntl.h>
> #include <unistd.h>
>
> using namespace std;
>
> typedef unsigned char u8;
> typedef unsigned int u32;
> typedef unsigned long long u64;
>
> struct e{
> u8* f;
> u64* s;
> };
> struct se{
> u64 cid;
> void* np;
> };
> const u64 a = -1;
> u32 subchunk = 2878;
> u32 subchunkperchunk = 140;
> u32 chunksize = 1028096; //in bytes
> u32 TSize = 4096;
> u32 eleOfChunk = 50;
> u32 eSize = 4096;
> u64 x;
> vector<e> v;
> int it=0;
>
> u64& add(const void* ep)
> {
> void* subChunkPtr = (void*) (((u64)ep) / chunksize * chunksize);
> u64 firstElementLocation = ((u64)subChunkPtr) + TSize;
> u32 indexInsideSubChunk = (((u64)ep)-firstElementLocation) / eSize;
> se* elePtr = ((se*)subChunkPtr) + indexInsideSubChunk;
> return elePtr->cid;
> }
>
> int main()
> {
> int file = open("Result", O_CREAT|O_WRONLY, S_IRWXU);
> dup2(file, 1);
> for (u32 i=0; i<= subchunk/subchunkperchunk; i++)
> {
> u32 eleInCurrChunk;
> if (i != subchunk/subchunkperchunk)
> {
> eleInCurrChunk = subchunkperchunk;
> }
> else
> {
> eleInCurrChunk = subchunk % subchunkperchunk;
> if (eleInCurrChunk == 0)
> {
> break;
> }
> }
> u32 CSize = (eleInCurrChunk+1) * chunksize;
> u8* sp;
> sp = new u8[CSize];
> u8* ssp = sp + (chunksize - ((u64)sp % chunksize));
> for (u32 j = 0; j < eleInCurrChunk; ++j) //creat block
> {
> u8* scp = ssp + j * chunksize;
> u8* ep = scp + TSize;
> for (u32 k=0; k < eleOfChunk ; ++k) //creat subblock
> {
> v.push_back(e());
> v[it].f = ep;
> v[it].s = &add(ep);
> add(ep)= a; //set the 64 bits to high
> ep += eSize;
> ++it;
> }
> system("free -m");
> }
> }
> return 0;
> }
>
>
