Re: interesting(?) data on network interrupt servicing

From: John Baldwin <jhb_at_freebsd.org>
Date: Thu, 23 Mar 2006 13:56:40 -0500
On Thursday 23 March 2006 12:50, Paolo Pisati wrote:
> On Thu, Mar 23, 2006 at 11:12:24AM -0500, John Baldwin wrote:
> > You probably want preemption on to minimize latency.  
> 
> i'm doing preemption now...
> 
> > If this is a
> > UP machine, you should turn SMP off.  It might be interesting to
> > compare using 7.x without APIC as well, since you are not using
> > APIC on 4.x.
> 
> ok, and what do we expect from it?
> besides interrupt masking/eoi, what are the other areas influenced
> by apic<->8259 switch?

That's probably about it.

> moreover, should i profile the asm part too?

Well, it would be good if you could do that on both 4.x and 7.0 to get
a better comparison.  4.x does most of the equivalent of
intr_execute_handlers() in asm.

> 
>        APIC                    8259
> 	|                       |
> 	|                       |
>      ISR_VEC()                INTR()         
> 	|                       |
>         |                       |             we don't take any measure
> ========================================== <= above this threshold...
>  lapic_handle_intr()    atpic_handle_intr()
>         \                    / 
>          \                  /
> 	intr_execute_handlers()
>                   |
> 		  |
> 		  .
> 		  .
> 		  . 

-- 
John Baldwin <jhb_at_FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org
Received on Thu Mar 23 2006 - 17:59:26 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:54 UTC