Re: page faults

>And please remember: When measuring page fault ratio, do so using a
>delta measure. Collect the walues, wait some time, collect another set
>of values and calculate the ratio of the difference between the two. The
>fact that the tcp/page fault ratio is increasing feels like you are
>measuring the ratio using absolute values, which gives a false figure.
>

Well yes that's what I have been doing! but using the delta rule shows no
big improvement on the values and they still show a ratio around 0.45 and
it's still increasing by time

>Another cause to pagefaults is if your machine runs out of free memory
>pages. I.e. squid is writing data faster than your OS sends the data to
>disks. What is the output of the free column of vmstat in relation to
>the values found in /prox/sys/vm/freepages multiplied by four?
>

vmstat shows :
        swpd: always 0
        free : always around 2200 (sometimes drops to around 1800 but very
rarely)

cat /proc/sys/vm/freepages shows:
        280 420 560
which shows that free never dropped below (420*4)=1680

free shows:
                total used free shared buffers
cached
Mem: 127936 125808 2128 38520 6848 5020
-/+ buffers/cache: 113940 13996
Swap: 100796 0 100796

from this I can see the system is doing no swapping, so where the pagefaults
values come from ??
Received on Thu Oct 29 1998 - 01:41:33 MST