Ideal value for buffer cache....

[vtpaulson]

Hi Friends,

I heard from my collegue that HP recommends keeping buffer cache at 300Mb, regardless of server size. This recommendation is based on the fact that overhead of maintaining the buffer cache linked list becomes larger than the benefit of buffering when it is over 300Mb.

Is this ture?? As a usual practice I increase the buffer cache as on when I hit the disk bottle neck. In some case its up to 50% of server memory. Max. value I have for buffer cache so far is 2GB. Do you think this doesn't make sense?

[Michael Tully]

Hi,

Have a look at these two links

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xba6146ff9277d511abcd0090277a778c,00.html

Doesn't matter on OS
http://us-support.external.hp.com/cki/bin/doc.pl/sid=ac9cfee90752a165a3/screen=ckiDisplayDocument?docId=200000049436468

HTH
-Michael

[A. Clay Stephenson]

The answer is that it depends. Typically servers dedicated to NFS benefit from large buffer caches but it most cases anything above 300 MB or so is at best of very little benefit and at worst harms. Although I tend to be in the minority, I tend to run fixed buffer caches by setting bufpages to a non-zero value.

The default max_dbc_pct value of 50% is almost always the first parameter than is changed to typically something in the 10-15% range unless you choose to set bufpages which defats the dynamic buffer cache. I suppose the best answer is to try it and see. In that case, it is much easier to measure using fixed buffer cache.

[Uday_S_Ankolekar · Accepted solution]

This topic has been discused many a times in this forum. The answwer is always it depends.!

The following thread might be helpfulhttp://forums.itrc.hp.com/cm/QuestionAnswer/1,11866,0x997cabe92dabd5118ff10090279cd0f9,00.html

Goodluck,
-USA..

[vtpaulson]

Guys,

Thanks for your answers...

By the way is there any direct way of measuring the buffer cache utilisation when we configured as dynamic buffer instead of setting a fixed value everytime to reach the optimum value??

I have a dedicated NFS server using TB's of EMC disks. In this case if I need to have more than 300MB of buffer cache is it advisable to put greater values? Then what kind of performance damage we are talking about if this is more than 300MB?

[Steven Sim Kok Leong]

Hi,

Use glance. gpm is easier to navigate around.

It is under
reports -> system information -> system table report.

Alternatively if you don't have glance, use the adb debugger to extract the current value.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com