From: Justin Willoughby (jwilloug@mhc.net)
Date: Thu Nov 20 2003 - 07:48:38 EST
Bill had the following reply (down a little ways) to a few questions I sent his way on this question.
It seems that "everything is not as it seems" .
I have a another system that I am guessing has a lot of memory devoted to buffer cache, but I don't think that it is shrinking when more memory is need.
A vendors application is crashing when it runs out of memory but the process is using less then 600MB of memory, and the whole system has 2GB of memory. Glance shows 100% memory utilization when the application crashes.
What I am wondering is, either way is the buffer cache not shrinking and/or why is swap not being utilized when the memory is at 100%? When the vendors application crashes I know swap is no where near being totally utilized. I am guessing I am just not understanding how HPUX is using memory and swap.
Thanks,
- Justin
==================
Justin Willoughby
Computer Operations
==================
>>> Bill Hassell <blhconsulting@mindspring.com> 11/10/03 04:58PM >>>
Yep, that's everything so it isn't meaningful to attribute
the usage to Ignite. You'll need to look at the more refined
memory stats (kernel, buffer cache, shared libraries, memory
mapped files, shared memory segments, user program data
space, etc). The buffer cache is very likely the primary user
so you can forget about it because it will shrink when other
processes need RAM.
Bill
The metric I have been using is GBL_MEM_UTIL which I believe shows the total memory used on the system, which is shown as a percent.
According to the book I have this metric is: "The percentage of physical memory in use during the interval. This includes system memory (occupied by the kernel), buffer cache and user memory.
- Justin
==================
Justin Willoughby
Computer Operations
==================
>>> Bill Hassell <blhconsulting@mindspring.com> 11/10/03 12:37PM >>>
Memory usage from Measureware can be several dozen metrics. It is
perfectly normal for a file-intensive app like Ignite to open and read
hundreds to thousands of files which will cause the buffer cache to
grow. That has nothing to do with Ignite. HP-UX grows the buffer
cache as needed until it hits the dbc_max_pct value or there are some
processes that need more memory and the cache will be reduced in
size. The actual appearance and behavior of the buffer cache is very
different between 11.00 and 11.11. There are dozens of reasons for
memory usage and Ignite is just demonstrating one i\of the complex
ways that HP-UX handles memory. Because it is getting kicked off
afterhours, the cache may not be used much until Ignite runs.
You'll need to look at measureware graphs for the various memory
areas like processes, shared memory, shared libraries, memory
mapped files, and buffer cache. Glance makes it easier to see memory
usage.
Bill
I graphed out memory utilization using data from measureware and can see a jump in memory shortly after Ignite starts but the memory never seems to go down after Ignite is done. I was pretty sure that it was Ignite as I don't believe I have anything else running at noon on Sunday. I will try and verify again that it is Ignite (by moving the time it starts and seeing if the jump follows the change in time).
--
Best regards,
Bill Hassell (blhconsulting@mindspring.com)
--
Best regards,
Bill Hassell (blhconsulting@mindspring.com)
--
---> Please post QUESTIONS and SUMMARIES only!! <---
To subscribe/unsubscribe to this list, contact majordomo@dutchworks.nl
Name: hpux-admin@dutchworks.nl Owner: owner-hpux-admin@dutchworks.nl
Archives: ftp.dutchworks.nl:/pub/digests/hpux-admin (FTP, browse only)
http://www.dutchworks.nl/htbin/hpsysadmin (Web, browse & search)
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 11:02:36 EDT