buffer cache hit ratio 100%

[Javed Khan_1]

Hi,

On one of my HP-UX system is observing

23:38:44 bread/s lread/s %rcache bwrit/s lwrit/s %wcache pread/s pwrit/s
23:38:45 0 71 100 5 8 38 0 0
23:38:46 0 13 100 2 2 0 0 0
23:38:47 0 4 100 1 1 0 0 0
23:38:48 0 2870 100 3 44 93 0 0
23:38:49 0 11275 100 3 82 96 0 0

Average 0 2835 100 3 27 90 0 0
$

Model : rp7420
OS : HP_UX 11.23

dbc_min_pct 5
dbc_max_pct 60

Total VM : 13.8gb Sys Mem : 3.1gb User Mem: 9.9gb Phys Mem : 18.0gb
Active VM: 13.6gb Buf Cache: 4.5gb Free Mem: 435mb FileCache: na


Can some one help me in getting buffer cache hit ratio to normal

Thanks in Advance

Regards,
Javed

[Vivek Bhatia]

Hi Javed,

can you paste the output of below commands?

#kcusage -d dbc_max_pct
#swapinfo -tam
#kcusage

Regards
Vivek

[Javed Khan_1]

$ /usr/sbin/swapinfo -mta
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 19500 211 19289 1% 0 - 1 /dev/vx/dsk/rootdg/swapvol
reserve - 10079 -10079
memory 18405 4586 13819 25%
total 37905 14876 23029 39% - 0 -
$


$ /usr/sbin/kcusage
Tunable Usage / Setting
=============================================
dbc_max_pct 24 / 60
maxdsiz 155119616 / 2063835136
maxdsiz_64bit 1118208 / 4294967296
maxfiles_lim 69 / 8192
maxssiz 1048576 / 134217728
maxssiz_64bit 28672 / 1073741824
maxtsiz 1363968 / 67108864
maxtsiz_64bit 16384 / 1073741824
maxuprc 80 / 14763
max_thread_proc 140 / 3000
maxvgs 0 / 256
msgmni 2 / 16404
msgseg 0 / 32767
msgtql 0 / 16404
nfile 1303 / 248108
nflocks 23 / 8192
ninode 1244 / 133280
nkthread 1224 / 28723
nproc 225 / 16404
npty 0 / 512
nstrpty 1 / 60
nstrtel 0 / 60
nswapdev 1 / 10
nswapfs 0 / 10
semmni 34 / 4096
semmns 46 / 16384
shmmax 34953688 / 5000000000
shmmni 13 / 400
shmseg 6 / 300
$

[Duncan Edmonstone]

So what this is telling you is that 100% of your filesystem reads are being satsified out of buffer cache rather than having to go to disk...

In what way do you think this is a *bad* thing?

> Can some one help me in getting buffer cache hit ratio to normal

What do you *think* a "normal" buffer cache hit ration should be????

HTH

Duncan

[James R. Ferguson]

Hi Javed:

To add to Duncan's comments, I'd say your system's performance must be damn good!

The question really is, are you having any performance issues? Are you end-users happy?

Regards!

...JRF...

[Steven E. Protter]

Shalom,

I would adjust dbc_max_pct to between 5-10.

Why:
1) A lot of memory is consumed by your current buffer cache settings and you may not be getting benefit. Oracle for example on 11.11 does not benefit at all from a buffer cache.
2) System performance suffers during times when the system is adjusting its buffer cache. Locking into a figure with dbc_min_pct and dbc_max_pct avoids this problem.

I liked to lock systems in at 5. Depending on how you use your system a different number might work better for you. But your current settings waste a lot of memory and impact your system when the system is adjusting itself.

SEP

[Bill Hassell]

AS mentioned, buffer cache read percentage of 100% is absolutely the best possible value. Your write cache percentage will always be lower, sometimes below 40-50%.

If you want to lower the percentage, run either (or both) of these two commands:

find / -type f -exec grep -q zzz {} \;
du -s /

These commands will overflow your buffer cache instantly and the percentage hit ratio will drop dramatically. Warning: This demonstrates how the buffer cache works and will definitely impact system performance. But it proves nothing. There is *NO* normal value for the percentage. The goal is 100% which you have already seen.

[Emil Velez]

buffer cache hit is 100% that is good but you only have 400MB of free memory.. Have you ever got low on memory where you system did paging.

If you did not get low on memory then your buffer cache is ok but what is running on that system. Could a application like oracle SGA benefit from being bigger while buffer cache is smaller.

[Javed Khan_1]

Thanks all

i want to free some amount of memory

[Javed Khan_1]

is there any standard set of kernel parameter which i should use for application like tomcat?

What are the best practices one should consider while installing tomcat on PA-RISC machines?

[Bill Hassell]

> i want to free some amount of memory

Other than reducing the dbc_max_pct kernel parameter, the only way to reduce memory usage is to stop running the big programs like databases. Or you can go to the DBA and ask them to reduce their shared memory usage. That will slow down performance but then you'll get additional memory. Is there a reason you have to reduce memory usage?

is there any standard set of kernel parameter which i should use for application like tomcat?

No, because it all depends on how Tomcat is being used. If you have a casual web page that is used every few minutes, there is nothing to change. If you have dozens of Java programs and threads presenting hundreds of web pages every second then you may need to adjust Apache values as well as open files (nfile) and processes (nproc), etc.