lkml.org 
[lkml]   [2000]   [Feb]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: latest 'guaranteed low latency' patch against 2.2.14

On Tue, 8 Feb 2000, Kevin Morgan wrote:

> MontaVista certainly won't recreate the work of Ingo! That is nice
> work. However, preemption points have their limitations, and are
> difficult to maintain over time. Also, preemption points don't
> address excessive interrupt off paths (which are significant and
> already reported as problems from more than one of our customers).
> [...]

how do you improve task-execution latency without rescheduling actually?
Short of adding new hardware, or without adding a new (limited) API to
execute user-specified code. Ie. how do you reschedule to a generic
user-task without actually rescheduling?

it's not hard at all to measure 'scheduling atom' latencies, and a runtime
check can be added as well using the Pentium cycle counter: if latency is
above a certain level then an alert is issued. Many RT OSs have watchdog
timers which guarantee application latency.

IRQ latencies are a completely orthogonal issue, and apart from SCSI IRQ
handler routines the latencies are well below 100 usecs. (apart from some
badly behaving driver) The SCSI latency can be helped by reducing the
amount of SCSI commands queued in a single IRQ instance.

> ps: yes all MontaVista product work in kernel space will of course be
> released under GPL, [...]

(great!)

> pps: We have no illusions about getting native Linux down the the
> RTLinux range of responsiveness, 15-20 microsecond "guaranteed" kinds
> of levels. RTLinux is a fine solution for that level of real-time
> requirement.

yep, completely agreed. Nevertheless there is a new method you might not
have heard about: the IOAPIC-based NMI watchdog i developed about a year
ago and which got added recently to the 2.3 SMP code. The NMI watchdog
also provides easy access to microsecond-resolution excuted (kernel level)
code - on standard PC (SMP) hardware. No kernel change is necessery, it's
part of the standard kernel. NMI interrupts are the 'secret backdoor' to
the PC architecture to guarantee hard-RT-latencies, even if the OS does
not directly support low latencies.

any device interrupt can be turned into an NMI interrupt - of course
special handlers have to be written, and careful coding is needed as no
other kernel facilities can be relied on.

-- mingo


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:56    [W:0.158 / U:0.284 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site