Re: How much memory is currently paged?

[DrFranken <midrange@xxxxxxxxxxxx>]

Let me toss in my $0.02 on this subject.

First This paging amount matters significantly in a couple cases. First 
if it's affecting the amount of available disk space. We have actually 
had to add disk to customer systems when the system is small and this 
number is large. The true fix was application fixes but the fast fix was 
more disk. Second is if this number never stop climbing. Again in this 
situation it's usually some sort of application issue but if it never 
stops climbing then clearly at some point you run out of disk. Third is 
if the amount of paging is significantly affecting jut just memory but 
adding to disk subsystem workload enough to matter there as well. And it 
CAN matter on small systems.

In your case you have a smaller amount paged out than you do total 
memory so by my personal 'rule of thumb' you are in the safe zone. I 
don't usually consider it a problem until paged memory exceeds 4x real 
memory. Even then if paging and especially faulting aren't problems then 
it's still a 'who cares' situation. To me it's faulting above all that 
tells the tale. If faulting is acceptable and wait-inel + act-inel == 0 
then you don't really have a problem.

How to fix it? You wrote the list. Sometimes merging pools is better 
than many small pools. Of course the issue there is task type. You don't 
want big SQL jobs running with your interactive people, wrong type of 
work but I've seen systems with 15, 20, more? memory pools from back in 
the day when they needed to assure this one batch program would respond 
quickly (did it myself even) but today that's much less of a problem. 
Sure auto-tune will move the memory around but often just about as it's 
no longer needed.

Keeping up with fixes for 'memory leaks' is important but I think you've 
applied a PTF or three in your day. I have seen some whoppers in this 
area over the years. Java has been a problem too in the past but I can't 
point to any issues in 6.1 or 7.1 with Java.

WAS and Domino and JAVA and the HTTP servers all abuse memory pretty bad 
at start-up. As you mention tuning for start-up isn't the normal setup, 
tune for run time and let the poor WAS guy get another cup of coffee 
while it cranks up.

    - Larry "DrFranken" Bolhuis

P.S. My personal favorite is "Add Memory" because I sell and install 
memory. :-)

On 8/13/2012 11:16 PM, rob@xxxxxxxxx wrote:
> There is a way to determine how much memory is paged.
> WRKSYSSTS
> Take
> Current unprotect used . :   110277 M
> Subtract the sum of:
>      Pool
>     Size M
>    5044.38
>   24317.51
>    2284.17
>   14762.66
>   14454.68
>    7480.57
> or
> 68343.97
> and you get
> 110277 - 68343.97
> =
> 41933.03
> or 41GB paged out.
>
> Then the million dollar question becomes:  What does this mean?
> Is there an acceptable amount of paging?  If so, is it an amount or a
> percentage?
> Or does this not matter, what really matters is paging and faulting?  If
> so, how does one address paging and faulting.
> - Shift memory from one pool to another.
> - Add memory
> - Change applications to use less memory
> - Terminate staff to reduce workload on system.
> - Move applications to other platforms.
> - Move back to paper based business from computer based business. (There's
> this ancient horror story from our distant past...)
> - and on and on and on...
>
> This system is a bit memory constrained at times and shifting memory from
> one pool to the next isn't always the answer as it quickly changes from
> one pool to the next.  For example, WAS startups often jump from hogging
> memory in base (where they run) and from interactive (where they are
> started from).  So please, no plugs for tuning products that are more
> responsive than QPFRADJ.
> Sure, startups shouldn't be your main concern (unless you're the IBM guy
> connecting in remotely at 11pm helping you get WAS stuff configured),
> since startups only occur, well, at startup.
>
> Sys      Pool   Reserved    Max  ----DB-----  --Non-DB---  Act-
> Pool    Size M   Size M     Act  Fault Pages  Fault Pages  Wait
>    1    5044.38   1300.66  +++++     .0    .0  165.4 442.7     .0
>    2   24317.51    183.91   1880     .0   2.7    2.7   5.4  55527
>    3    2284.17       .48     28     .0    .0     .0    .0   1090
>    4   14762.66       .60   1547     .0    .0    6.3   6.3   1254
>    5   14454.68       .12    157     .0    .0     .0    .0     .0
>    6    7480.57      4.57    800     .0    .0     .0    .0  26454
>
>
> Rob Berendt