Top and RAM information

[GRANDIERE]

I have problem to understand the top memory information. I have too servers HP-L2000 with the same version of HP-UX and the same RAM.

On the first one, top returns that good memory information :
Memory: 2628368K (1851980K) real, 5655072K (3811928K) virtual, 13396K free

On the seconde it returns that bad information :
Memory: 126004K (72696K) real, 131932K (97608K) virtual, 5732K free

The second server has already 2 Gb of RAM.

What is the problem with that second top ? How can i be sure of top informations.

Thanks for help,

Regards,

Dahu

[Muthukumar_5 · Accepted solution]

Check with sam -> performance monitor -> system properties -> memory tab for memory informations for two machines?

Is it correct for two machines?

hth.

[Pedro Cirne]

Hi,

Please post the output of "dmesg"

Thks,

Pedro

[Joseph Loo]

hi,

i do not think u should use top if u r trying to gather system information/statistics.

for physical memory, u may use sam as described by Muthu or:

# dmesg|grep Physical
for total physical memory or

also, if u r running make_tape_recovery, run print_manifest command for other hardware, software and kernel info.

regards.

[Yogeeraj_1]

hi,

try
echo "selclass qualifier memory;info;wait;infolog"|cstm

regards
yogeeraj

[GRANDIERE]

The first one $dmesg | grep Physical :

Physical: 3145728 Kbytes, lockable: 2366116 Kbytes, available: 2721264 Kbytes

The second server :
Responds nothing.

Thanks.

[Joseph Loo]

hi,

for the first server, u have

for the second, dmesg has been spooling other messages thus the memory info is gone. use sam -> Performance Monitor -> System Properties -> Memory tab

or what yogi has stated:

echo "selclass qualifier memory;info;wait;infolog"|cstm

to find the physical memory.

regards.

[Mahesh Kumar Malik]

Hi Dahu

Memory information from top command is never accurate since it is result of incidental probe. To get exact memory use use sam -> Performance Monitor -> System Properties -> Memory option

Regards
Mahesh

[GRANDIERE]

On the first server sam says :
Physical Memory: 3074.8 MB
Real Memory:
Active: 1706416.2 KB
Total: 2646925.3 KB
Virtual Memory:
Active: 3667562.3 KB
Total: 5813675.5 KB
Free Memory Pages: 3055 at 4 KB/page

On the second sam says :
Physical Memory: 2049.8 MB
Real Memory:
Active: 70135.5 KB
Total: 119637.6 KB
Virtual Memory:
Active: 95670.9 KB
Total: 133323.8 KB
Free Memory Pages: 3235 at 4 KB/page

My first server has performance problem. It has 7 instances of ORACLE. And i wonder if the Memory is not full. Thats why i try to compare with the second server. I am suprised to see that they have quite the same quantity of free memory pages.

What a pity : i wanted to prouve that we have too mutch instances of ORACLE.

Regards,

Dahu.

[Joseph Loo]

hi,

u may like to run the following command to check the memory statistics:

# vmstat 3 20
virtual memory stat twenty 3 seconds interval.

would also like to know how much of SGA have u given for each instance:

# ipcs -mbop
should be able to view the memory allocation of each instance. or u may try a select statement to find the SGA size.

more info tomorrow, good night.

regards.

[A. Clay Stephenson]

Top is a meaningless tool for this task. Top only concerns itself with memory that is directly associated with processes. It knows nothing about kernel data structures like buffer cache. Using top to determine the amount of memory (free or otherwise) in a system is much like looking at your car's temperature gauge to determine how much coolant is in the system. The quantities are related but they aren't the same.

[GRANDIERE]

hi,

Good night ... see u tomorrow ...

ORACLE SGA parameters :
db_block_size = 8192
db_cache_size = 268435456
shared_pool_size = 134217728

See in attachement à file with the informations you asked.

I stopped one of my 7 ORACLE instances dans the server is more performed.

Thanks,
Regards,

Dahu.

[Joseph Loo]

good morning,

looking at your attachment, there r actually two/three instance taking up 740261888 bytes and four/five instances taking up 824152064 bytes, based on the SEGSZ column from ipcs output.

as u have stated that there r only 3GB in that system, the 6 instances currently running is already taking whatever memory u have left.

u have a few alternatives:

1) lower the SGA size of the instance which may affect performance
2) shutdown instances if not in use and this box is not production?
3) consider use swapping but i am really against that
4) BUY more memory if your box still have empty memory slots.

when u may run vmstat over a period when u r shutting down each instance, u will notice a increase in the free memory column.

for info, to determine the sga size for each instance:

SQL> show sga

regards.

[Ted Buis]

Time to check some kernel parameters:
dbc_max_pct
buf_pages (if it exists)
swapmem_on
swchunk
max_swap_chunks (if it exists)

Also what version of HP-UX are you running?
(uname -a)

The two systems don't appear to have the same RAM, so you may want to use stm or xstm to see exactly what is loaded. Maybe they were not loaded properly or there is bad RAM.

[GRANDIERE]

Hello,

>/etc #uname -a
HP-UX loiret B.11.11 U 9000/800 137901527 unlimited-user license

That's my Stupide question : how print HP-UX parmeters you asked ?

Thanks.

Dahu.

[Kent Ostby]

/usr/lbin/sysadm/system_prep -s

Also, make sure that you have at least 2xRAM worth of device swap as reported from swapinfo -t

[GRANDIERE]

You'll find the system_prep results in the attached file. I can't see the parameters you asked.

Here is the swap info :
swapinfo -t
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 6291456 489196 5802260 8% 0 - 1 /dev/vg00/lvol2
reserve - 5086388 -5086388
memory 2370364 148424 2221940 6%
total 8661820 5724008 2937812 66% - 0 -

[Ted Buis]

You can view kernel parameter in via 'sam'. Or 'kmtune' ('kctune' is the new command) with no parameters should also list all kernel parameter settings.

[GRANDIERE]

Hi,

Here are the parameters you asked for :
dbc_max_pct = 50
buf_pages = 0
swapmem_on = 1
swchunk = 2048
max_swap_chunks = doesn't exists

Thanks,
Dahu.

[Yogeeraj_1]

hi Dahu,

do you have good reasons for having so many instances?

consider using different schema in only a few instances...

regards
yogeeraj

regards
yogeeraj

[Ted Buis]

Your dbc_max_pct is probably too high. I don't see any reason to go above 25% and possibly 20% on the system with 2GB and 15% on the system with 3GB RAM. Are the parameters the same on both systems?
You have more memory (at least more that is working) and more virtual memory on one system than the other. Do you have OnLineJFS? If so, you should check out this read this cookbook to see how to avoid double buffering with Oracle.

[Ted Buis]

By the way, I think it is just swap_chunks, rather than max_swap_chucks, but it is the maximum value. But we don't need it. It would be large enough by default in 11.11 for your amount of RAM. Also, since you have pseudoswap turned on, memory can all be used regardless of your physical disk swap size.

[GRANDIERE]

Hi,

I work with developpers for decreasing the number of ORACLE instances. They will have to manage the shema information in their programmes instead of use the ORACLE_SID.

My system knowledges are not enougth to manipulaite system parameters. I dont want to modify theses one.

The last interresting information you coult say me would be a command to know if my system swaps too much. Is that information easy to collecte ?

Thank a lot for help,

By,

Dahu.

[Ted Buis]

The page out information is found in the output from vmstat you did earlier. It is in the po column. The is the important part of "swap" which is really a misnomer. You need to examine it during periods of heavy usage, to see how much this is a problem. Your output didn't show much activity, but it was a small sample.

Changing kernel parameters can be easily done in sam, but you may need to do a reboot for them to take effect.

[Yogeeraj_1]

hi again,

another tip that may help you is something that we do at our site where we only have ONE database instance and many schema each associated with a specific environment and where we use the following statement to switch:

e.g.
execute immediate 'alter session set current_schema=scott';

kind regards
yogeeraj