dbc_max_pct at 100%

[Speedware]

I just want to know if its normal that the Kernel parameter dbc_max_pct's percentage of use is always 100%. With Kcweb, the percentage of use is showing to be 100% and never going down.

By the way, I am having this problem, on Itanium box, Hp 11.22.

[Steven E. Protter]

No.

We set dbc_max_pct to 7 dbc_min_pct to 5

If these two figures are too far apart, its very expensive in CPU terms to change the system. Very bad for performance.

More on performance.

http://www1.itrc.hp.com/service/cki/search.do?category=c0&docType=Security&docType=Patch&docType=EngineerNotes&docType=BugReports&docType=Hardware&docType=ReferenceMaterials&docType=ThirdParty&searchString=UPERFKBAN00000726&search.y=8&search.x=28&mode=id&admit=-682735245+1092253190882+28353475&searchCrit=allwords

SEP

[Dave Olker · Accepted solution]

Hi,

Are you saying that dbc_max_pct is really configured at 100%, or is your question is it normal for the buffer cache to sit fixed at the dbc_max_pct size and not shrink?

According to kctune:

Tunable dbc_max_pct
Description Maximum percentage of memory to be used for caching file I/O
data and metadata
Module fs
Current Value 50 [Default]
Value at Next Boot 50 [Default]
Value at Last Boot 50
Default Value 50
Constraints dbc_max_pct >= 1
dbc_max_pct <= 90
dbc_max_pct >= dbc_min_pct
Can Change Immediately or at Next Boot


This parameter cannot be sized above 90% of physical memory, so my guess is your question is whether it's normal for the buffer cache to consume dbc_max_pct of memory and stay there.

If that is your question, the answer is yes, it is not unusual to see this on a system that is using buffer cache resources on a regular basis and is not otherwise memory constrained.

It's my understanding that buffer cache will grow the dbc_max_pct as buffer cache needs grow. The only time it will get sized back down is if the system comes under memory pressure and vhand starts stealing pages from buffer cache. Otherwise the memory will stay allocated to the buffer cache.

Also, just because the buffer cache has allocated all dbc_max_pct of physical memory, that doesn't necessarily mean that all of those buffers are actually in use at that time.

Hope this helps,

Dave

[Jeff Schussele]

Hi,

On a moderately to heavily used box it's entirely common for the buffer cache usage (which is what dbc_max_pct defines) to be 100% used. The OS is going to use what you give it after all.
NOW - where that might become a problem is when the value - which is a % of total RAM - is set *way* too high. Like the default of 50%.
So don't worry about the usage - worry about the actual kernel parameter value. If it's still at the default, bring it down to a value that will result in a max RAM value of 400 - 800 MB. That seems to be the sweet spot, any more can be a waste.

HTH,
Jeff

[Speedware]

Thanks
That answer my questions. Right now its set to the default 50.