It used to be my understanding that top's definition of memory "use" was the number of physical pages mapped to any processes. However, recently I saw something that made me think otherwise. I'd really like someone to explain.
As an example of what I'd expect given my old understanding, below the totaled vmsize from /proc/ is almost three times the "used" in top (89916452KB, 32787948KB). Some pages are mapped to more than one process, some pages are mapped to disk (and so shouldn't be counted at all by top), etc.
$ grep VmSize /proc/*/status | awk '{total += $2} END{print total}'
89916452
$ top
top - 09:39:34 up 5 days, 7:37, 9 users, load average: 4.89, 4.14, 3.92
Tasks: 242 total, 2 running, 239 sleeping, 0 stopped, 1 zombie
Cpu(s): 69.0%us, 0.5%sy, 0.0%ni, 30.3%id, 0.0%wa, 0.0%hi, 0.2%si, 0.0%st
Mem: 32961916k total, 32787948k used, 173968k free, 248132k buffers
Swap: 16777208k total, 1584300k used, 15192908k free, 7659572k cached
However, I recently came across the following. The totaled vmsize is smaller by almost a factor of 4! How could this possibly be the case? Even if one were to argue that the kernel pages that are mapped into each process aren't counted in the VmSize entry in /proc/*/status, one would still have a tough time making the argument that they account for any sizable fraction of the ~46GB in use.
$ grep VmSize /proc/*/status | awk '{total += $2} END{print total}'
8519308
$ top
top - 09:43:55 up 291 days, 15:16, 8 users, load average: 1.02, 1.08, 1.08
Tasks: 291 total, 1 running, 290 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.0%us, 0.0%sy, 0.0%ni,100.0%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 49324520k total, 46017056k used, 3307464k free, 5448168k buffers
Swap: 8388600k total, 228k used, 8388372k free, 36486236k cached
Please explain.
