memory question

[mike worrell]

what is the best way to check and see if
you have a memory issue??

swapinfo shows only 23% being used and also
it shows lvol2 has 3%
Is this good or a problem?
Trying to troubleshoot a problem, any help
would be appreciated.
Thanx
Mike

[Vincenzo Restuccia]

SAM->performance monitors->system properties

[Robin Wakefield]

I always have a look at vmstat output, under the "sr" column (always ignore the 1st line of figures). This tells you how many pages are being scanned in order to free up memory for use be other files/processes. If you set vmstat to monitor every 30 seconds, and this column is always zero, then you're OK. If it's regularly over 200 in this timeslice, you're definitely short of memory.

Robin

[mike worrell]

thanx Robin,
I am at zero ,

[Jim Turner]

Mike,

Start Glance and let's do some looking around.

First, look at the memory utilization bar graph at the top. Is it above 98%? How much is System(S), User(U), and Dynamic Buffer Cache(B)? System should be a relatively small portion. If not, you've got some app(s) going ape sh*t with too many syscalls. User should be the majority of memory usage. If buffer cache is taking a big chunk, you need to make a kernel change. HP-UX out of the box will have dbc_min_pct set to 5 and dbc_max_pct set to 50. With today's large memory models, someone at HP was smoking crack when they chose the 50% number. It has been my experience that cache hit rates seem to reach a sweet spot at around 300MB of dbc. I've actually seen hit rates diminish when given an enormous cache. I recommend calculating what percentage 300MB is of your total RAM and hardcoding both min and max_dbc_pct to this number.

Now let's look at the Memory Report (m) in Glance. Scan down the row headings to the bottom where you see "VM Reads" and "VM Writes". Watch the "Current" column for these two metrics as your samples go by. Are you seeing *consistent* VM Reads and Writes? If you've tuned your dbc (from above) and eliminated any possible app(s) that were running amok (also referenced above) and you are *STILL* seeing > 98% memory utilization and consistent VM Reads and Writes, it's time to buy some more RAM, my friend.

You pay a horrendous price in performance for swapping to disk. Swap is a good safety valve to be (actively) used once in a while, but using it consistently really buggars up your response times. BTW, don't panic if you see swap space that is reserved. Swap reservation doesn't cost you anything in terms of performance.

The last possibility is that you have high memory utilization but *no* consistent VM Reads and Writes. This simply means that you're getting your money's worth out of your RAM. Nothing more and nothing less. If you expect growth in your application(s), you might consider buying more RAM to add as needed, but there is no need of immediate panic.

I hope I have helped. Sorry if I told you things you already knew.

All The Best,
Jim