Re: High interrupt CPU usage in top

[Mark Cullen <mark.r.cullen@xxxxxxxxx>]

Matthew Dillon wrote:
> :Well, world and kernel just finished compiling, without INVARIANTS, and 
> :there's no difference. Infact, it seems worse... but it's probably the 
> :same. I guess I should keep INVARIANTS in for debugging in case of panics?
> :
> :http://img198.imageshack.us/img198/1571/interrupt3eg.jpg
> :
> :40% and very few interrupts? That was just starting out compiling 
> :`mysql4-server`.
>
>     INVARIANTS is pretty light weight, it won't make much of a difference.
>
>     Try doing some basic tests to exercise JUST the network adapter, like
>     running a 'ping -i 0.1 <target>'.  And some tests to exercise JUST your
>     hard disk, like doing a few cpdup's or a tar or something.
>
>     If its the disk, check your dmesg output to whether the ATA driver is
>     using DMA or not.
>
> 					-Matt
> 					Matthew Dillon 
> 					<dillon@xxxxxxxxxxxxx>

Well, `ping -i 0.1 <t>` got me less than 1%. However, `ping -f <t>` got 
me 35%.

As for disk, a very simple `dd if=/dev/<d> of=/dev/null` gets me ~25%, 
and as far as I can see DMA is enabled.

---
(root@bone2)/usr/pkg/etc atacontrol mode 0
Master = ???
Slave  = WDMA2
(root@bone2)/usr/pkg/etc atacontrol mode 1
Master = ???
Slave  = UDMA33
---

As you can see, a bit of a mix up there, CD drive is primary slave and 
the hard disk is secondary slave. However, before you blame it on 
this... my last install (just yesterday actually) was a 30GB disk on 
primary master and has the exact same behaviour.

I ran the same commands on the 4.11 box (different motherboard / chipset 
but that's it) and `ping -f <t>` is around 17% and the simple dd is the 
same, +- a few percent. Both have polling enabled, and both are using 
identical Intel NIC's, two in each machine.

Any of that info is worth anything to anyone?