ninode mem usage seen in kmeminfo output

[Dan Copeland]

What bucket would ninode cache fall into -- sample output attached.

tia,
Dan

[Jeff Schussele]

Hi Dan,

Although I'm not 100% sure, I believe it's in the kernel-text/static bucket.
It's defined by the vx_ninode kernel parameter which is quite frequently overlooked when tuning the kernel to minimize kernel memory requirements. It defaults to 0 which causes the usage to be determined by total memory installed in the system. Which could cause the value generated to be several hundred thousand on systems with lots of memory. And that's a waste of kernel memory.
Realistically I've *never* seen a system that needed more than 40,000.

Rgds,
Jeff

[Scot Bean · Accepted solution]

The HFS inode cache is in static system memory. See

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B2355-60103/B2355-60103_top.html&con=/hpux/onlinedocs/B2355-60103/00/41/4120-con.html&toc=/hpux/onlinedocs/B2355-60103/00/41/4120-toc.html&searchterms=ninode&queryid=20041208-130841

Why do you need to know, are getting "inode table is full" error msgs?

[Dan Copeland]

being told to reduce ninode kernel parm to save system memory...just trying to make sure it's even worth doing.

ninode is set pretty high (40000)

Dan

[Scot Bean]

Try 'bdf -t hfs' to see how many HFS filesystems you have.

"The value of ninode can be lowered on systems with small or no mounted HFS file systems, to reduce memory consumption."

[Dan Copeland]

i believe that nfs also contributes to this usage -- that's where we get some usage.

Dan

[Scot Bean]

I would agree about NFS, but only if the NFS files being shared are on HFS filesystems.

Since 11.0, most folks use HFS only for /stand and if this is true on your machine, I would say 40000 is way too high.

[Scot Bean]

Try to see how many of your HFS inodes are in use. 2 ways:

- sar -v 1 1

- Glance, under pulldown menu
Reports, System Info, System Tables Report

[Scot Bean]

Here's a popular whitepaper on kernel tuning:

http://h21007.www2.hp.com/dspp/ddl/ddl_Download_File_TRX/1,1249,280,00.pdf

It DOES reflect your concern about cross-impacts against the NFS Client cache.