Page Size Successes/Failures on vmstat

[Chris Cruz_1]

I ran a vmstat -s on our L class server, and it listed all these successes/failures...what do they mean?


4291368 Page Select Size Failures for Page size 16K
1136 Page Select Size Failures for Page size 64K
1141 Page Select Size Failures for Page size 256K
1144 Page Select Size Failures for Page size 1M
1148 Page Select Size Failures for Page size 4M
15274492 Page Allocate Successes for Page size 4K
2065301 Page Allocate Successes for Page size 16K
141 Page Allocate Successes for Page size 64K
117 Page Allocate Successes for Page size 256K
69 Page Allocate Successes for Page size 1M
9 Page Allocate Successes for Page size 4M
2 Page Allocate Successes for Page size 16M
9 Page Allocate Successes for Page size 64M

[Jeff Schussele · Accepted solution]

HI Chris,

These are requests from all processes that are requesting a *specific* memory page size over a specific time.
As the request size gets larger the chance of failure grows due to memory fragmentation.
It could, but not necessarily, indicate memory pressure. But I always use all the other tools like glance & Measurware/PerfView to verify memory bottlenck as well as good old response times before I bump up RAM.

HTH,
Jeff

[Arunvijai_4]

# man vmstat says,

-S Report the number of processes swapped in and out (si and so) instead of page reclaims and address translation faults (re and at).

interval Display successive lines which are summaries over the last interval seconds. If interval is zero, the output is displayed once only. If the -d option is specified,
the column headers are repeated. If -d is omitted, the column headers are not repeated.

The command vmstat 5 prints what the system is doing
every five seconds. This is a good choice of printing
interval since this is how often some of the statistics
are sampled in the system; others vary every second.

count Repeat the summary statistics count times. If count is
omitted or zero, the output is repeated until an
interrupt or quit signal is received. From the
terminal, these are commonly ^C and ^\, respectively
(see stty(1)).

[Leonardo Guzmán Wideman]

Hi.

Running 'man vmstat' I can see a capital -S option and a non capital -s option.
Chris asked for the second.
Arunvijai answered for the first.

The man page segment says the next:

-s Print the total number of several kinds of paging-related events from the kernel sum structure that have occurred since boot-up or since vmstat was last executed with the -z option.

Running the same line command on an hp-ux 11.11, I got the next:

# vmstat -s
318913 swap ins
319102 swap outs
786 pages swapped in
64137 pages swapped out
387164473 total address trans. faults taken
219021734 page ins
370418 page outs
113118727 pages paged in
2799005 pages paged out
139040770 reclaims from free list
108803925 total page reclaims
11283 intransit blocking page faults
2165045635 zero fill pages created
339085275 zero fill page faults
135527698 executable fill pages created
2093825 executable fill page faults
0 swap text pages found in free list
14751058 inode text pages found in free list
163 revolutions of the clock hand
32496660 pages scanned for page out
423148 pages freed by the clock daemon
9391489033 cpu context switches
8042960107 device interrupts
67576076014 traps
750404850142 system calls
24309793 Page Select Size Successes for Page size 4K
14113857 Page Select Size Successes for Page size 16K
11303021 Page Select Size Successes for Page size 64K
740372 Page Select Size Successes for Page size 256K
154463 Page Select Size Successes for Page size 1M
22462 Page Select Size Successes for Page size 4M
5102 Page Select Size Successes for Page size 16M
626 Page Select Size Successes for Page size 64M
24309793 Page Select Size Failures for Page size 16K
36266244 Page Select Size Failures for Page size 64K
25066022 Page Select Size Failures for Page size 256K
25801543 Page Select Size Failures for Page size 1M
25955948 Page Select Size Failures for Page size 4M
25967961 Page Select Size Failures for Page size 16M
25973063 Page Select Size Failures for Page size 64M
321555213 Page Allocate Successes for Page size 4K
25574198 Page Allocate Successes for Page size 16K
14255442 Page Allocate Successes for Page size 64K
402324 Page Allocate Successes for Page size 256K
76479 Page Allocate Successes for Page size 1M
22701 Page Allocate Successes for Page size 4M
5529 Page Allocate Successes for Page size 16M
186 Page Allocate Successes for Page size 64M
1044723 Page Allocate Failures for Page size 4K
232228 Page Allocate Failures for Page size 16K
116101 Page Allocate Failures for Page size 64K
22457 Page Allocate Failures for Page size 256K
28907 Page Allocate Failures for Page size 1M
32077 Page Allocate Failures for Page size 4M
32733 Page Allocate Failures for Page size 16M
33666 Page Allocate Failures for Page size 64M
345846 Page Demotions for Page size 16K
3152636 Page Demotions for Page size 64K
166505 Page Demotions for Page size 256K
2482 Page Demotions for Page size 1M
478 Page Demotions for Page size 4M
522 Page Demotions for Page size 16M
122 Page Demotions for Page size 64M
#

Regards.

[Bill Hassell]

Prior to the PA-8000 processor chip set, the memory page size was always 4K. Once the PA-8000 was introduced, HP-UX was modified to accomodate a larger page size. One of the advanatages of a larger page size is that the limited number of (very high speed) mapping registers could point to a larger section of memory. If a very large program bounced all over program memory, the small 4K page registers would have to be reloaded a significant number of times. By making the page larger, these register faults and reloads c an be minimized.

So for programs that have been compiled/chatr'ed for variable page size, the kernel will try to accomodate the page requests and the above are stats. Now whether these stats are good, average or terrible depend very much on the measurement length. vmstat keeps adding until either a reboot or vmstat -z is used to zero the stats.

Normally, I would not be too concerned unless you have many large (compute-bound) programs running that may be running slower than expected. The page-fault rate (aka, TLB or page register reloads) is an indication of smaller than optimal page sizes. It will take a fairly detailed knowledge of how the programs work to determine whether any fix is necessary.