topic Re: Physical memory usage of kernel in Operating System - HP-UX

Physical memory usage of kernel

Achilles_2 — Thu, 15 May 2008 11:22:20 GMT

Hi there,

I have two rx4640, with install 16GB of Physical memory. The same OS version but the physical memory usage of kernel are different via kmeminfo command.

Server 1 :
=========
Physical memory usage summary (in page/byte/percent):

Physical memory = 4189222 16.0g 100%
Free memory = 1795653 6.8g 43%
User processes = 1682435 6.4g 40% details with -user
System = 695314 2.7g 17%
Kernel = 485852 1.9g 12% kernel text and data
Dynamic Arenas = 199482 779.2m 5% details with -arena
vx_inode_cache = 43323 169.2m 1%
vx_global_pool = 34866 136.2m 1%
spinlock = 24388 95.3m 1%
vx_buffer_cache = 17888 69.9m 0%
vm_pfn2v_arena = 16566 64.7m 0%
Other arenas = 62451 243.9m 1% details with -arena
Super page pool = 20010 78.2m 0% details with -kas
Static Tables = 200994 785.1m 5% details with -static
pfdat = 98184 383.5m 2%
nbuf = 45824 179.0m 1% bufcache headers
vhpt = 32768 128.0m 1%
bufhash = 8192 32.0m 0% bufcache hash headers
text = 7389 28.9m 0% vmunix text section
Other tables = 8636 33.7m 0% details with -static
Buffer cache = 209462 818.2m 5% details with -bufcache

Server 2:
=========
Physical memory usage summary (in page/byte/percent):

Physical memory = 2092068 8.0g 100%
Free memory = 923798 3.5g 44%
User processes = 779572 3.0g 37% details with -user
System = 375567 1.4g 18%
Kernel = 270963 1.0g 13% kernel text and data
Dynamic Arenas = 82796 323.4m 4% details with -arena
vx_global_pool = 18910 73.9m 1%
spinlock = 11721 45.8m 1%
vm_pfn2v_arena = 8406 32.8m 0%
VFD_BT_NODE = 5242 20.5m 0%
vx_inode_cache = 4246 16.6m 0%
Other arenas = 34271 133.9m 2% details with -arena
Super page pool = 24050 93.9m 1% details with -kas
Static Tables = 119341 466.2m 6% details with -static
pfdat = 49032 191.5m 2%
nbuf = 35184 137.4m 2% bufcache headers
vhpt = 16384 64.0m 1%
text = 7544 29.5m 0% vmunix text section
bufhash = 4096 16.0m 0% bufcache hash headers
Other tables = 7100 27.7m 0% details with -static
Buffer cache = 104604 408.6m 5% details with -bufcache

How come the kernel text and data in server 1 greater than server 2?

I hope that you can help me.

Thanks

Re: Physical memory usage of kernel

Achilles_2 — Thu, 15 May 2008 11:28:33 GMT

sorry, server 2 has 8GB and server 1 has 16GB

Re: Physical memory usage of kernel

James R. Ferguson — Thu, 15 May 2008 11:33:27 GMT

HI:

Fisrt, it would seem that server-1 has 16GB physical memory whereas server-2 has 8GB.

Regardless, server-1 shows about 12% kernel text and data whereas server-2 shows 13%.

Do you really consider that a difference? I certainly don't. I presume too, that these servers were booted at different times and have undergone different work loads since. You aren't comparing fairly from that standpoint, then, either.

Regards!

...JRF...

Re: Physical memory usage of kernel

Don Morris_1 — Thu, 15 May 2008 12:52:45 GMT

Text? All I can think is a different driver module on one vs. another. The tiny difference you're talking about would only entice me to track it down if I truly had nothing better to do with my life.

Data -- Well, first there's the data that is memory descriptor (pfn-to-virt entries, physical frame descriptors (pfdats)). Those are allocated per memory page... so more memory pages, more of them. (Hence they stay a flat percentage cost but you're trying to look at the raw costs.)

Other than that -- the rest is dynamic data differences. Dynamic being dynamic... it would be a function of the load. I really don't see anything odd here.

Re: Physical memory usage of kernel

Tim Nelson — Thu, 15 May 2008 13:29:52 GMT

As Don mentioned there are a number of kernel configs that are dynamic in nature and allocated at time of boot. Examples that comes to mind is the min_dbc_pct/max_dbc_pct another example is vx_ninode. I am sure there are 100s of others.

So unless you hard code every one of these your mileage will vary. Especially when % of 16GB or 8GB are involved.