Laymans answer for Average Memory and Free Memory

[Belinda Dermody]

When I see the results from vmstat as below, I need a LAYMENS ANSWER what the avm and free refers to. I have a 7450 with 2 850 Cpu's and 1gig of memory installed.. Am I getting to a situation where I need to increase memory.

avm free
150895 1551
150895 1462
153695 1756
153695 1639
152667 1787
152667 1772

[RAC_1]

avm is active vitrual memory. (virtual memory is swap space configured + ram)
free mememory is free mememory available.

Note that the values are in pages. One page is 4096k

Anil

[Eric Antunes · Accepted solution]

Hi James,

You will obtain similar information about VM with:

#swapinfo -tam

And if you run:

#vmstat -dnS 6 1000

You will see how much processes are swapped (w processes in procs)

Best Regards,

Eric Antunes

[Belinda Dermody]

I am looking for an explanation of the NUMBERS in layman terms so I can explain it to my manager(and myself, man pages and reading isn't my best area of work, I need black and white explanation to understand things) and if we need to purchase additonal memory.

[Eric Antunes]

In swapinfo command I refered above, you get the USED and FREE memory in Mb...

TYPE AVAIL USED FREE USED ...
...
memory 1129 228 901 20%
...

[Geoff Wild]

To determine if you need increase memory, you need to check other things besides vmstat.

Things like:

GBL_MEM_PAGEOUT_RATE
GBL_MEM_SWAPOUT_RATE
GBL_MEM_QUEUE

You can get that in glance and measureware.

Rgds...Geoff

[Pete Randall]

What it means is that you've got roughly 150,000 pages of memory in use by processes and between 1500 and 1700 pages still available - not in use.


Pete

[Bill Hassell]

Here's the answer for your manager: free memory in vmstat is not meaningful because HP-UX uses virtual memory. So your 1Gb of RAM can run a *lot* more than 1Gb of processes. I have personally run programs totalling 35Gb in just 1Gb of RAM. (oh, did I mention that the system response time was measured in minutes?)

vmstat will show how HP-UX keeps readjusting the processes and data areas in memory so that the most amount of work an be done. But once you exceed all of your RAM, the system does not stop (like other less capable systems) but starts using swap space. But swapping requires stopping preocesses, then swapping memory to and from disk and restarting processes. This is no problem if the end user is on lunch break but for active programs, the performance penalty can be very high.

The *only* useful column in vmstat is po (Page Outs). General rules for po:

0-9 no problems
10-99 some slowdown, RAM is a bit small
100-up major speed and response problems (delays logging in, etc) RAM is massively too small.

SO vmstat is the tool for memory sizing, just ignore avm and free, and look at po. You'll need to at least double or triple RAM size if your po column is 3 digits for long periods.

[Eric Antunes]

Hi,

vmstat also shows you what are the
processes states (vmstat -nS):

procs
r b w
4 0 0

You need to maximize b processes and avoid b (blocked for resources like I/O) and w processes (swapped). In the example above I've 4 processes in run queue and none blocked for resources or swapped...

Best Regards,

Eric Antunes

[Eric Antunes]

Where I wrote "You need to maximize b processes and avoid b (blocked for resources like I/O) and w processes (swapped). In the example above I've 4 processes in run queue and none blocked for resources or swapped... ", I wanted to say:

You need to maximize r processes and avoid b (blocked for resources like I/O) and w processes (swapped). In the example above I've 4 processes in run queue and none blocked for resources or swapped...