mathematics of swap space

Herbert van den Bergh hbergh at nl.oracle.com
Tue Apr 23 22:47:35 AEST 1991


In article <2758 at sapwdf.UUCP> Bill Wohler <wohler at sap-ag.de> writes:
>swap space gurus:
>
>  with 128M of real memory and 208M of swap (and effective total of
>  336M of memory!) on a 530, i would not have expected to see "INIT:
>  swap space low", especially considering when the sum of the SIZE
>  column of "ps uax" is only 58M.
>
>	ps uax | awk '{sum += $5}
>	END {print sum}'
>	57876
You're adding the SZ column here. The Performance Monitoring and
Tuning Guide descibes this column as:

SZ	The virtual size of the process. Specifically, SZ is the
	amount of paging space used by the data segment of the
	process, plus the amount of paging space used by the text
	segment of the process, expressed in kilobytes. The text
	segment of the process is generally resident on the
	filesystem, and thus does not use pages from paging space.

	*Note:* SZ values are subject to error in the first release of
	the system.

Don't ask me what "first release" means. That's probably what we all
have now.

>
>  researching this problem via info revealed the following numbers:
>
>[wohler at aix3:175]% lsps -a 
>Page Space  Physical Volume   Volume Group    Size   %Used  Active Auto
>paging00    hdisk1            uservg          64MB      65     yes yes
>hd6         hdisk0            rootvg         144MB      34     yes yes
>					     ---
>					     208
This shows that (0.65*64M + 0.34*144M) = 41.60 + 48.96 = 90.56MB is in use.
>
>[root at aix3:201]# vmstat
>procs    memory             page              faults        cpu     
>----- ----------- ------------------------ ------------ -----------
> r  b   avm   fre  re  pi  po  fr   sr  cy  in   sy  cs us sy id wa 
> 1  0 24401 11756   0   0   0   0    2   0 117   55  37  1  1 96  0
>
>
>24401 (avm) * 4096 = 96604k
>
>11756 (fre) * 4096 = 47024k
>	           --------
>                    144628k
>
>From the same document:

avm	This data is not represented as a rate. It is the number of
	active virtual memory pages at the time of the interval
	sample. It represents the number of page space pages being
	consumed at that instant.

fre	This data is not represented as a rate. It is the number of
	real memory pages on the free list at the time of interval
	sample.

So the avm column of vmstat is pretty close to the lsps percentage in
use.

>  these numbers and info raise the following questions which info
>  couldn't answer.  i'd be very thankful for any of your answers:
>
>	o what percentage of paging space free is considered to be a
>	  low condition (ie. SIGDANGER's sent to processes)?

Maybe one of you IBM guys can answer this one?

>	o is the total amount of available virtual space just the swap space,
>	  or real memory plus swap space?

See the above description of avm: it stands for ACTIVE virtual memory.

>	o what program is used to determine how much real memory the
>	  system thinks it has?

One way of doing it is "lscfg -l mem"

>	o why is the swapper process so large?

Because it shows the amount of paging space used by the kernel
segment. The ps option u seems to have a bug that doubles this size
for the swapper. "ps vg" shows a SIZE column for swapper that is half
the size of the "ps aux" SZ column.

>	o why is the total amount of available and free memory
>	  displayed by vmstat not equal to 208 displayed by lsps?

See the description of avm and fre, they don't mean what you think
they mean.

>	o why is there such a discrepancy between the sum of the SIZE column
>	  of ps and the avm field of vmstat?

See the *Note:* at the end of the SZ description.

>  in summary,  do these numbers indicate a bug or a broken
>  configuration?  why did the system indicate that we ran out of swap
>  space?  what other tools do we have at our disposal?

You can use 'sar -r' to see how many paging space pages are still
free while your processes are running.

>-- 
>						--bw
>-----
>Bill Wohler <wohler at sap-ag.de> <sapwdf!wohler>
>Heidelberg Red Barons Ultimate Frisbee Team


-- 
Herbert van den Bergh,		Email:	hbergh at oracle.nl, hbergh at oracle.com
ORACLE Europe
--



More information about the Comp.unix.aix mailing list