7

As many have stated on StackExchange and elsewhere, a typical kernel/mmu page size for Linux is 4 KiB. I have verified that this is the case on my OS/architecture (RHEL 6.6, Intel Xeon E5-2690).

Now among other things, the vmstat -s command reports on "pages paged in" and "pages paged out", which are counters that start from 0 at boot time. I have confirmed that at any given moment, these values are exactly equal to the values found for the pgpgin and pgpgout entries in /proc/vmstat (does vmstat command get them from /proc/vmstat?). Other commands, and in particular sar -B report pgpgin/s and pgpgout/s as Kibibytes paged in/out per second during a particular interval.

In recent tests, I have seen that pgpgin/s and pgpgout/s values from sar also correspond exactly to a rate calculated from vmstat values for a specified interval. This leads me to conclude that vmstat is reporting values for pages of 1 KiB in size. Thus the pgpgin/out values of /proc/vmstat are the number of KiB paged in/out since boot.

My questions are:

  1. Is this a valid conclusion, and
  2. If so, why do vmstat and other facilities report in 1 KiB pages instead of the 4 KiB pages that are the "common currency" of the OS and architecture? In particular, vmstat is saying "pages paged in/out", not "KiBs paged in/out". This is confusing.
1
  • Side note: 4 KiB is a ridiculously small page size, so I expect it to become less and less "common".
    – kubanczyk
    Dec 6, 2016 at 9:21

1 Answer 1

4

In regards to the counters in /proc/vmstat the function that appears to update these values is submit_bio in Linux/block/blk-core.c.

2067 blk_qc_t submit_bio(struct bio *bio)
2068 {
...
2079                         count = bio_sectors(bio);
2080 
2081                 if (op_is_write(bio_op(bio))) {
2082                         count_vm_events(PGPGOUT, count);
2083                 } else {
2084                         task_io_account_read(bio->bi_iter.bi_size);
2085                         count_vm_events(PGPGIN, count);
2086                 }
...
2100 }
2101 EXPORT_SYMBOL(submit_bio);

bio_sectors is a macro in Linux/include/linux/bio.h

 64 #define bio_sectors(bio)        ((bio)->bi_iter.bi_size >> 9)

This seems to suggest to me that the numbers are counted in sector sizes (512) bytes. Given thats what value you'd end up getting if you right shifted a byte count by nine.

This appears to be a similar setup between kernels 2.6.27 - 4.0.x anyways.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .