Buffer header 100% utilization, is this normal?

[jane zhang]

Hi all,
The system table report of glanceplus (gpm) shows that the buffer header(nbuf) is 100% used as follows,
System Table Avail Used Util% High%
buffer headers(nbuf) 6167 6167 100% 100%

our system is working fine, Should we turn this kernal parameter?

Thank you,
Jane

[Michael Tully]

Hi,

The first thing I would look at is how much
buffer cache you have in the kernel. These
days having any more than 300Mb may cause
symptoms such as this. What is the current
parameter for 'nbuf', 'dbc_max_pct' and
'dbc_min_pct' ? Also how much RAM do you
have ?

-Michael

[S.K. Chan]

I suggest leaving nbuf and bufpages at 0. Instead tweak
DBC_MAX_PCT = 50% (default)
DBC_MIN_PCT = 5% (default)
My guess is you have nbuf and bufpages defined to a value (non zero) without really knowing how much is needed.

[Steven Gillard_2]

The systems buffer cache is always kept full, so what you are seeing is normal. There is a 'least recently used' algorithm to replace buffers in this cache when new file system reads occur.

Regards,
Steve

[jane zhang]

Hi all,
Thanks for the response.
I used sam to check the kernal parameter, nbuf 0
bufpages 7392
dbc_min_pct 2
dbc_max_pct 2
I also use /sbin/dmesg to check the available RAM
available: 1864580 Kbytes
We have Oracle RDBMS installed on this HP-UX B.11.00 U 9000/800.
How can I know which algorithm is used in our system? or LRU is universal to all HP-UX OS?


Regards,

Jane

[Sandip Ghosh]

In this scenario I would suggest to change as follws

Bufpages from 7392 to 0
dbc_max_pct from 2% to 15%
dbc_min_pct from 2% to 5%

Sandip

[Sandip Ghosh]

Can you please post a output of sar -b 5 10

Sandip

[jane zhang]

Here is the report.
Thanks,

# sar -b 5 10

HP-UX fcdsjus B.11.00 U 9000/800 03/21/02

11:07:00 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
11:07:05 0 41 100 1 6 79 0 0
11:07:10 0 20 100 2 6 62 0 0
11:07:15 0 1 100 2 5 62 0 0
11:07:20 0 28 100 1 8 84 0 0
11:07:25 0 37 100 3 6 46 0 0
11:07:30 0 108 100 4 5 33 0 0
11:07:35 0 35 100 7 11 35 0 0
11:07:40 0 22 100 5 9 40 0 0
11:07:45 0 10 100 3 6 52 0 0
11:07:50 0 28 100 3 8 64 0 0

Average 0 33 100 3 7 55 0 0

[Sandip Ghosh]

I don't think your system is active right now. Because it is showing no activity. Collect this report when the Server is most active. Look at the %rcache. Remember you have to bring this column between 90 to 100. That will be your optimum setting. You can achieve this by increasing/decreasing the buffer area. Right now there is no activity , that is why you are getting 100% as %rcache.

If it is still 100% in the busiest time you need not to change the dbc_max_pct.

Sandip

[Roger Baptiste]

hi,

Since you are using dynamic buffer cache, the nbuf row in glance should be showing utilization as N/A . Are you sure glance is ok on the box.

-raj

[Frank Slootweg]

Sandip wrote:

> bufpages from 7392 to 0
> dbc_max_pct from 2% to 15%
> dbc_min_pct from 2% to 5%

Just to explain this:

If you want Dynamic Buffer Cache (i.e. set the dbc_*_pct parameters), then *both* nbuf *and* bufpages *must* be 0 (zero).

If nbuf or bufpages (or both) are non-zero, then you have a Fixed (size) Buffer Cache and the settings of the dbc_*_pct parameters are irrelevant.

As far as I know, the SAM Help explains this rather well.