SuSE 10 32bit Hardware interupts on VMWare ESX 4i

Linux and Unix-Like SuSE
1

Hi!

Does anyone know how a high percentage of Hardware Interrupts does affect a Linux system? any why does it generate hardware interrupts?

The affected system is a SuSE 10.2 32bit running 2.6.16.60-0.33-vmi #1 SMP Fri Oct 31 14:24:07 UTC 2008 i686 i686 i386 GNU/Linux

procinfo shows that the vmxnet driver has the highest HI.

Linux 2.6.16.60-0.33-vmi (geeko@buildhost) (gcc 4.1.2 20070115) #1 SMP Fri Oct 31 14:24:07 UTC 2008 4CPU [xxxxxx]

Memory:      Total        Used        Free      Shared     Buffers      
Mem:       3107720     2128564      979156           0      259596
Swap:      1049560         108     1049452

Bootup: Sun Jan 22 16:51:02 2012    Load average: 0.03 0.01 0.00 2/122 25043

user  :       1:52:53.69   0.2%  page in :     647174  disk 1:    30843r 3580099w
nice  :       0:00:10.87   0.0%  page out:   52266616
system:       1:04:58.18   0.1%  page act:    8754422
IOwait:       0:32:11.18   0.0%  page dea:    1116251
hw irq:      16:38:12.93   1.8%  page flt: 1520224988
sw irq:       0:01:18.77   0.0%  swap in :          0
idle  :  37d 18:50:29.63  97.8%  swap out:         27
uptime:   9d 15:50:29.72         context :  713058193

irq  0:        88 VMI-alarm             irq  9:         0 acpi                 
irq  1:         8 i8042                 irq 12:       114 i8042                
irq  3:         1                       irq 14:   7496633 ide0                 
irq  4:         1                       irq 51:  12623625 vmxnet ether         
irq  6:         5                       irq 59:   2399766 ioc0                 
irq  7:         0 parport0              irq 67:         0 vmci                 
irq  8:         2 rtc

4 Cores
3GB RAM

2

Hardware interrupts are how hardware devices(or "hardware" devices, in a VM) signal the system that data is ready, or a condition needs to be dealt with. The more network traffic, the more interrupts for vmxnet; if the system runs long enough they can be very high indeed. Nothing wrong with it. It's certainly better than the alternative -- polling!

3

Ok thank you for the reply, I'm looking deeper into this.