vx-ninode again

[derek b smith_1]

I know there are threads out there and I have read them. But after reading them and after reading "Memory Usage - Whats is using all of the memory?" By: Eric Herberholz at HP and HP Perf cookbook I am confused on what to set this to.

Eric recommends setting vx_ninode to 20k - 30k as a start. I want my first setting to be permanent for a long time! After writing this quick script on my production system :
bdf -i |grep -v "/stand" |awk '{print $6}' |grep -v iused |awk '{sum+=$1;printf("sum: %d\n",sum)}'

I get a total in use value of 234340. Is this the value I want to use to trend my future setting other than the default 0?

thx a bunch!

derek : )

[Kevin Wright · Accepted solution]

no. I would set it to 30K, or 2% of mem. (1 inode= about 2K I think)

echo vxfs_ninode/D | adb -k /stand/vmunix /dev/mem

[Jeff Schussele]

Hi derek,

I concur.
The default value of 0 can lead to values of more than 1/4 million on large RAM systems.
I've been advised by experts that one *never* needs more than 40k in real life.
This is definitely one of those parms that *should* be tuned properly out-of-the-box, but isn't.

My $0.02,
Jeff

[derek b smith_1]

ok guys thx for the responses, but I want to make for certain that this change will not effect production. This value 234340 represents all of my JFS file systems iused sum from bdf -i. So everyone is telling me that 20k - 30k is sufficient for vx_ninode
even though when I echo this parameter I get

echo "vxfs_ninode /D" |adb -k /stand/vmunix /dev/kmem

240640?????

why is 20-30k recommended considering this number above and has anyone been in the situation before? has anyone changed it w/out negative results?

thank you,

[Jeff Schussele]

Hi derek,

The number drawn from the kernel is the *max* number of open vx inode table entries allowed. So think about just how *many* files you'll ever have open. I doubt you'll ever have more than 40k at any one time. So why not save the kernel memory & set the value to a realistic value?

Rgds,
Jeff

[derek b smith_1]

yes I definitely agree with "I doubt you'll ever have more than 40k at any one time. So why not save the kernel memory & set the value to a realistic value?"


So the 234340 number can be ignored and taken for granted? Is the number in any way useful?

[Jeff Schussele]

Well first understand that this is a *cache* table so it's not an "in use" table - it's a "used" table. That number simply reflects the *total* number of cache entries used since the last boot. What's most important however IS the in use value - i.e. how many entries are currently being used by all processes running *right now*?
Consider 2 things:
A) What's the memory "cost" of the cache table size?
B) What's the CPU "cost" of searching an unnecessarily large table?
Those "costs" are what I'm pointing out. You don't need a huge table for exactly those two reasons. Set it to 40k & forget it.

My 2 cents,
Jeff

[derek b smith_1]

I will set it to 40k, but IF it is a "number that simply reflects the *total* number of cache entries used since the last boot"
then why does it flucuate?

[derek b smith_1]

In one of my 11.11 systems this parameter vx_ninode does not exist. What do I have to load or do to correct this?

[Kevin Wright]

Derek, you have received some valuable information for free, and you assigned points to 0 out of 47 replies!!!!!!!!!!!!!!!