Hi David,
What you are saying is true. But the
Point which I am trying to convey is the cost of the
context switch to the kernel thread "softirqd". From
the code I get that the scheduling of softirqd by the scheduler
is still "Non-deterministic" as the kernel is non-preemptive.
And most router are based on Deterministic data & control path,
using preemptive kernel. So perfromance-wise the "Non-deterministic"
behaviour is still an issue if we want to use linux for "Core Router "
applications unless Real-time extensions are used.
Cheers
----- Original Message -----
From: "David S. Miller" <davem@redhat.com>
To: <sureshsingh.keisam@analog.com>
Cc: <kevin.curtis@farsite.co.uk>; <linux-net@vger.kernel.org>
Sent: Thursday, November 14, 2002 9:32 AM
Subject: Re: possible delays in netif_rx
> From: "Suresh Singh K." <sureshsingh.keisam@analog.com>
> Date: Wed, 13 Nov 2002 20:27:26 +0530
>
> Next time when the kernel thread "sofirqd" is scheduled , it executes
> the RX softirq. As the version you are using is a non-preemptive
kernel,
> the control paths are all non-deterministic. So the longer interrupt
> processing
>
> Absolutely wrong, softirqs run at the return from any hardware
> interrupt trap as long as we are not already inside of a hw
> interrupt.
>
> softirqs runs to handle softirqs in process context when softirqs are
> "overloaded".
>
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
