strange top list ( HPUX 11.11 L3000)

[Xavier BLANC]

hello
( sorry search function doesn't reply so ... )
a sample produced by top!
CPU TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
0 ? 21271 userx 148 20 743M 7056K sleep 2:55 46.95 46.87 process1
0 ? 19949 root 148 20 893M 10472K sleep 5:39 5.94 5.93 process2
1 ? 7010 root 154 20 889M 6576K sleep 1:11 5.88 5.87 process3
1 ? 29566 root 154 20 900M 16956K sleep 8:04 5.33 5.32 process4
0 ? 20101 root 154 20 894M 11508K sleep 23:46 4.49 4.49 process5
1 ? 20833 root 154 20 891M 7984K sleep 4:01 3.55 3.54 process6

Two questions
first : SIZE are very high isn't it ? Is it normal ?

second question : the PRI value change when the process become in "run" state.
I was thinking that you could change priority with nice when you start the process. A process can be from 154 to 236


Of course we are experimenting perf problem so we are investigating.
Thank's in advance

[John Poff]

Hi,

The size depends on the process. Some processes will reserve much more memory than others (Oracle, for example), so you really have to look at each process and figure out what it is doing and what is normal.

The priority will go up and down as the process moves on and off of the CPU. It is how the scheduler keeps track of which process is due to run next.

What specific performance problems are you seeing on the system?

JP

[RAC_1]

A look at top gives feeling that those proceess are waiting for some resource. If you have glance you can chwck the global wait states as follows.

glance -B
You can also have a look at a particular process to know what it is waiting for. Also what kind of a bottlneck you are experiencing?? Mem, CPU, network??

Anil

[Jeff Schussele]

Hi,

The smaller size - under "RES" - is the actual memory footprint at the time of the snapshot. The larger value would be the absolute max size IF all the process's stack, text and data were memory resident. Something that rarely if ever happens. But that IS the size that will be reserved in swap space....just in case.

As to priority it's true that it will shift ALL the time. When you run nice on it all you do is shift somewhat how those changes occur.
When process PRI is between 128 <=> 153 they are in kernel range & generally are sleeping waiting on I/O and are unsignallable.

HTH,
Jeff

[Steve Steel]

Hi

As indicated . What is high . What are the processes

Have you change dthe KErnel parameters


Look at
ftp://eh:spear9@hprc.external.hp.com/memory.htm


Will give a lot of ways to check memory usage


Steve Steel

[Xavier BLANC]

Thank for you all
We are using Progress DB these process are executable that run progress code this, could be like this on other customer machine (hpux11.00):
? 29005 root 168 20 7496K 1404K run 62:16 0.05 0.05 _progres

These process are in sleep state that's not the problem (I hope! I can provide full top listing but .. is it worth).

- Oracle process server are huge in memory these are program's for gui interface. not Db server.

- glance -B not found

We can explain some problem's as we have a disk bottleneck: we have to spread a db over several disk.

But when disk activity isn't high we still have this phenomen : imagine a list 10 first ligne ok, then stop, then ten next lignes, then stop ... ( I really have to improve my english )

Our system is L3000 4 Gb ram 4 36Gb ) 12 external disk (fiber channel), everything mirored ( 2int and 6ext) .
120 users upon 91D Progress DB.

The network is ok !

Ok i'm reading ftp://eh:spear9@hprc.external.hp.com/memory.htm

Where should we start
Regards

[D Block 2]

you might find a few books helpful:
there are some new books recently published around hpux certification..


what is you free swap ?
# swapinfo -tam


best to use pseudo-memory, verify 16384
# kmtune |grep maxswapchunks
maxswapchunks 16384

what is your ORacle SGA for shared-memory
# ipcs -a

get the GLANCE command, a free 30 period installed.

here's some misc:


UNIX95= ps -e -o vsz=Kbytes -o ruser -o pid,args=Command-Line | sort -rkn1
UNIX95= ps -el -o pid,comm,sz|grep -i -v pid|sort -nr -k 3,3


# sar collection

sar -o /tmp/sar.data 10 360
# monitor every 10 secs, for total of 260 times or a 24 hour day
sar -f /tmp/sar.data



MONITORING BUFFER CACHE: p. 481 HP Certified Book

sar -b 1 5
bread/s - read ops per sec from Disk to BufferCache
lread/s - read ops per sce from buffer cache
%rcache - buffer cache hit ratio fro read requests
bwrit/s - write ops/sec from buffer cache to disk
lwrite/s - write ops/sec to buffer cache
%wcache - buffer cache hit ratio for write requests
pread/s - read ops/sec from raw device
pwrit/s - write ops/sec to raw device
sar -b 1 5

HP-UX B.11.11 U 9000/800 10/18/04

14:39:35 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s


--

[Xavier BLANC]

Comming back to top and it's values. Explain me this !
output from top :
1 ? 5604 userx241 20 663M 24408K run 44:21 82.14 82.00 _progres
output from :#UNIX95= ps -e -o vsz=Kbytes -o ruser -o pid,args=Command-Line | sort -rkn1| grep 5604
32032 userx 5604 _progres -pf session.pf

top say : 663M ps say 32032 Kb ?????
I'm totaly lost ...

Thank you tom
here's my swpinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 377 3719 9% 0 - 1 /dev/vg00/lvol2
reserve - 1621 -1621
memory 3153 2099 1054 67%
total 7249 4097 3152 57% - 0 -


sar -b 1 5
HP-UX serv1 B.11.11 U 9000/800 10/27/04

10:42:32 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
10:42:33 16 1333 99 276 395 30 0 0
10:42:34 13 587 98 279 273 0 0 0
10:42:35 16 2579 99 309 372 17 0 0
10:42:36 38 1987 98 324 337 4 0 0
10:42:37 146 1228 88 239 405 41 0 0

Average 46 1541 97 285 356 20 0 0

This morning (french hour) the user seems to be ok with the perf : measure is no phone call :)

I don't understand the note with
" maxswapchunks 16384 " my wish it "no swaping" if I must buy ram, we'll do it, ram is cheap against human time.

[Dietmar Konermann · Accepted solution]

The "SIZE" metric of top(1) is *not* the same as "vsz" of ps(1), as you already noticed. The ps(1) command only adds up data, stack and text sizes, while top(1) also includes the process' share of shared resoures.

If e.g. a process is one of ten being attached to a 1GB shmem segment, then you will see its share (here 100MB) with top(1), but not with ps(1).

Best regards...
Dietmar.