vx_ninode and sar

[Bryan D. Quinn]

Can anyone enlighten me as to how to monitor my vxfs_ninode utilization? I know that if I do sar -v I can see my inod-sz, but this reflects my ninode kernel setting, not the vx_ninode setting. If I do echo 'vxfs_ninode/D' | adb /stand/vmunix /dev/kmem that verifies what my kernel param is set at, but how do I monitor how close I am getting to that setting. Something comparable to sar -v for the ninode setting.

Any thoughts, suggestions or corrections would be welcomed.

As always, Thanks!

-Bryan

[Brian Bergstrand]

AFAIK, sar does not query this kernel var. It would be nice if it did. Glance may help, but I can't remember for sure. And of course, Glance isn't free.

[RAC_1]

May be putting it in a script and chekcing at regular intervals.

Something like

while true
do
echo vx_inode/D|adb -k /stand/vmunix /dev/kem
sleep 10
done

[T G Manikandan · Accepted solution]

check the script here

http://www2.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000062922494

[Stefan Farrelly]

You need the kmeminfo command, this displays details on memory including vx_ninode;

Kernel dynamic memory usage (in page/byte/percent):
Dynamic = 235396 919.5m 11% Kernel dynamic memory
Arenas = 158610 619.6m 8% Kernel arenas
vx_buffer_cache = 55520 216.9m 3%
vx_global_pool = 32264 126.0m 2%
vx_inode_cache = 11347 44.3m 1%
M_TEMP = 7430 29.0m 0%
ALLOCB_MBLK_LM = 6758 26.4m 0%
VFD_BT_NODE = 5659 22.1m 0%
M_REG = 4481 17.5m 0%
M_PREG = 4296 16.8m 0%
M_SWAP = 4212 16.5m 0%
M_SPINLOCK = 3888 15.2m 0%
KMEM_ALLOC = 3198 12.5m 0%
M_IOSYS = 2977 11.6m 0%
M_LVM = 2963 11.6m 0%
ALLOCB_MBLK_DA = 1476 5.8m 0%
M_KTHREAD = 1377 5.4m 0%
Other = 10764 42.0m 1% Other arenas...
Kalloc = 76731 299.7m 4% kalloc()
SuperPagePool = 60178 235.1m 3% Kernel superpage cache
BufcacheBufs = 9472 37.0m 0% Buffer cache bufs
BufcacheHash = 5120 20.0m 0% Buffer cache hash heads
Other = 1961 7.7m 0% Other...
Eqalloc = 55 220.0k 0% eqalloc()


Basically, the guideline is to set vx_ninode to 90% of ninode - no higher. If you leave set to 0 it can continually expand to even use Gigabytes of ram! we had this problem until we changed it to a non zero value.

Ive tried zipping kmeminfo and attaching - lets hope it works....

[Pete Randall]

Bryan,

Try "bdf -i".


Pete

[Stefan Farrelly]

attachment didnt work..email me and I will email you a copy unless someone knows where to downoad it ?

stefan_farrelly@ipcmedia.com

[Steven Gillard_2]

This will give you what you want:

echo "vx_cur_inodes/D" | adb /stand/vmunix /dev/kmem

Note that this is only an inode cache, so its going to be close to full most of the time. Thats not a problem.

Regards,
Steve

[Bryan D. Quinn]

Thanks everyone!

TG and Stefan for the script and kmeminfo.

Pete,

How I could have missed a simple solution is beyond me? Some times the KISS method slips right passed me!

Thanks as usual for your help!

Thanks again everyone!

-Bryan

[Bill Hassell]

Just to clarify, vx_ninode (like ninode) is a cache but the metrics about % full are very misleading. vx-ninode is handled differently than ninode. ninode contains current and recently opened inode information in order to save disk searches for commonly used files. However, ninode contains entries for HFS files only (such as /stand) so it should not be made very large (perhaps 1000 to 4000 max). However, ninode's size influences the DNLC, vx_ninode and ncdnode, so it is a good idea to get Dave Olker's book on HP-UX NFS which is the best explanation of how all these parameters influence each other,

[Steven E. Protter]

We've had a similar experience to Stephen Farelly concerning vx_inode.

Its critical to have its value as non-zero.

I'm posting in a link to a good tuning document.
http://www2.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=-1335382922+1063290834746+28353475&searchCrit=allwords

Doc ID: UPERFKBAN00000726

SEP

[Bryan D. Quinn]

Thanks Bill,

I knew I should have gotten that book. I sat through his NFS presentation at HP World and then had the chance to go buy the book and have him sign it! I could kick myself. Oh well, Amazon here I come.

Thanks for the clarification. By the way, I was only able to attend one of your sessions at HPWorld, because I was only there for two days. It was the 'HP-UX 11.x Features and Technology' session, enjoyed it very much thanks!

Looking forward to your book, when it comes out.

-Bryan