Hard Page Faults...

[Peter Clarke]

Hi

When i do a show system or if i do a show process this show's me the total number of page faults for this process.This is fine but is there any way of finding out how many of these are hard page faults and what this process is doing when the hard faults are occuring???

Peter

[Wim Van den Wyngaert]

To my knowledge :
. not with lexicals
. not with show commands
. not with sda show commands

So leaves complicated sda (for the experts with a pointy hat) and performance advisor.
"Monitor page" can report you the hardfaults too (read I/O) and much more.

Wim

[Willem Grooters]

I think you'll need MONITOR to distinguish between hard- and softfaults.
To find which process is involved most, you'll need to record and analyze afterwards, or have several (not too many) sessions running different MONITOR classes.
MONITOR SYSTEM shows a little bar in the MEMORY display, that gives the percentage of hard faults (left) en soft faults (right).
MONITOR PROCESSS/TOPFAULT will show you the top-faulting process - that includes hard AND soft faults.
MONITOR PAGE gives you the over-all view, with a distinction of hard and softfaults.

I think you will not find a direct relation between a process and hard faults. The swapper process takes care of hard faults, not the process.

Willem

[John Eerenberg]

> This is fine but is there any way of
> finding out how many of these are hard page
> faults?

SDA can do it for a given process. The location is at PHD + phd$l_pgfltio.
For example using a pid of 000003d3

$ analaze/system
SDA> set process/index=3d3
SDA> sho proc
SDA> exa phd + phd$l_pgfltio

Repeat the last SDA command and you'll see it increment if hard faults are occuring for a given process. Time the hard faults and you'll have a process's hard fault rate.

> what this process is doing when the hard
> faults are occuring?

To what level of detail?


For starters, setup two additional terminal sessions.

For the second terminal sessions (keep the one above so the fault rate is easy to access), go into SDA and so the following:
$ analaze/system
SDA> set process/index=3d3
SDA> sho proc/chan
You'll see what files are open.

In a third terminal session, do a show process.
$ show process/contin/id=3d3

Between these three terminal sessions, you'll get some idea of what is going on. And if you can use the PC from the show process and you have the source code, you'll be able to draw some conclusions, well, hopefully anyway. :-)

Keep in mind that when an image begins execution, it has to pagefault which includes some number of hardfaults. These are what I call good hard faults; of course too many hard faults and they are bad ones.

Figuring out what hard faults are good and when they become bad is whole 'nother topic. ;-)

enjoy,
john