latency report

latency report

[Trevor Woerner]

I have been performing an investigation to determine the suitability
of using GNU/Linux as a basis for embedded real-time (ish) work and
have prepared a report of my findings. I have made the assumption that
the application logic will be coded as a number of user-space
applications which interact with the hardware. As such, interrupt
latency alone is of little interest; instead I focused more on
measuring the time it takes some data to travel into the target
device, through a bunch of user-space processes (connected via named
FIFOs), then back out the device.

My report currently compares the stock kernel.org 2.6.14 with Ingo's
2.6.14-rt2 patches on two target boards. All preemption settings have
been evaluated. A full explanation of how these tests were conducted,
the various components that were used, the test scenarios that were
created, the results I measured, kernel configurations, and required
patches are included in the report.

This project's index can be found here ("Linux Latency"):
http://geek.vtnet.ca/embedded.html

All of the tools, scripts, and code I wrote to measure and generate
this report can be found in the project tarball (~1.7MB bzip2):
http://geek.vtnet.ca/embedded/LatencyTests/latencytests-3.2.1.tar.bz2

The report is here in html:
http://geek.vtnet.ca/embedded/LatencyTests/html/index.html

Code documentation (in various formats) and a PDF of the report are
also available.

Ingo Molnar has been incredibly helpful and provided an enormous
amount of initial feedback and suggestions; thank you.

Please CC me on any replies.
Thank you.

Re: latency report

[Gene Heskett]

On Friday 04 November 2005 02:06, Trevor Woerner wrote:
>I have been performing an investigation to determine the suitability
>of using GNU/Linux as a basis for embedded real-time (ish) work and
>have prepared a report of my findings. I have made the assumption that
[...]
>
>The report is here in html:
>http://geek.vtnet.ca/embedded/LatencyTests/html/index.html

And while the index displays properly, all links, when clicked on, lead
to local files which of course in html, do not exist on my machine.

-- 
Cheers, Gene
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
99.35% setiathome rank, not too shabby for a WV hillbilly
Free OpenDocument reader/writer/converter download:
http://www.openoffice.org
Yahoo.com and AOL/TW attorneys please note, additions to the above
message by Gene Heskett are:
Copyright 2005 by Maurice Eugene Heskett, all rights reserved.

Re: latency report

[Trevor Woerner]

On 11/4/05, Gene Heskett <gene.heskett@verizon.net> wrote:
> >The report is here in html:
> >http://geek.vtnet.ca/embedded/LatencyTests/html/index.html
>
> And while the index displays properly, all links, when clicked on, lead
> to local files which of course in html, do not exist on my machine.

I'm currently at work (i.e. miles and miles away from my home machine
and outside my network) and I can read through the report just fine.
Other people seem to be able to view the report fine (from both the
comments I've received and the entries in my log files).

Which links in particular are giving you problems?

Re: latency report

[Gene Heskett]

On Friday 04 November 2005 09:44, Trevor Woerner wrote:
>On 11/4/05, Gene Heskett <gene.heskett@verizon.net> wrote:
>> >The report is here in html:
>> >http://geek.vtnet.ca/embedded/LatencyTests/html/index.html
>>
>> And while the index displays properly, all links, when clicked on,
>> lead to local files which of course in html, do not exist on my
>> machine.
>
>I'm currently at work (i.e. miles and miles away from my home machine
>and outside my network) and I can read through the report just fine.
>Other people seem to be able to view the report fine (from both the
>comments I've received and the entries in my log files).
>
>Which links in particular are giving you problems?

Anything clicked on while displaying the above address in FF-1.07.

-- 
Cheers, Gene
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
99.35% setiathome rank, not too shabby for a WV hillbilly
Free OpenDocument reader/writer/converter download:
http://www.openoffice.org
Yahoo.com and AOL/TW attorneys please note, additions to the above
message by Gene Heskett are:
Copyright 2005 by Maurice Eugene Heskett, all rights reserved.

Re: latency report

[Randy.Dunlap]

On Fri, 4 Nov 2005, Gene Heskett wrote:

> On Friday 04 November 2005 09:44, Trevor Woerner wrote:
> >On 11/4/05, Gene Heskett <gene.heskett@verizon.net> wrote:
> >> >The report is here in html:
> >> >http://geek.vtnet.ca/embedded/LatencyTests/html/index.html
> >>
> >> And while the index displays properly, all links, when clicked on,
> >> lead to local files which of course in html, do not exist on my
> >> machine.
> >
> >I'm currently at work (i.e. miles and miles away from my home machine
> >and outside my network) and I can read through the report just fine.
> >Other people seem to be able to view the report fine (from both the
> >comments I've received and the entries in my log files).
> >
> >Which links in particular are giving you problems?
>
> Anything clicked on while displaying the above address in FF-1.07.

I don't know what that means, but I tried and followed
about half the links successfully.

-- 
~Randy

Re: latency report

[Carlos Antunes]

On 11/4/05, Trevor Woerner <twoerner.k@gmail.com> wrote:
>
> My report currently compares the stock kernel.org 2.6.14 with Ingo's
> 2.6.14-rt2 patches on two target boards.
>

Trevor,

rt2 was seriously screwed up (I posted about the problems in here).
rt3 solved the problems. You might want to repeat your testing with at
least rt3.

Carlos

--
"We hold [...] that all men are created equal; that they are
endowed [...] with certain inalienable rights; that among
these are life, liberty, and the pursuit of happiness"
        -- Thomas Jefferson

Re: latency report

[Trevor Woerner]

On 11/4/05, Carlos Antunes <cmantunes@gmail.com> wrote:
> rt2 was seriously screwed up (I posted about the problems in here).
> rt3 solved the problems. You might want to repeat your testing with at
> least rt3.

-rt3 lasted for all of, what, 30 minutes before -rt4 came out? :-)

I tested -rt5 last night and still have problems. I'm trying out -rt6
now. I was trying to isolate the problem but I haven't had any luck
yet.

If I boot a machine with 2.6.14-rt5 running either preemption k3 or
k4, simply running my "dolots.sh" script (which uses my "drone"
program, both of which are in the project package) will cause an OOM
problem after about 15 minutes or so. Running preemption setting k1
all night I found it was still happily running (and quite responsive
even with a loadavg of ~70) this morning.

Increasing the load to upwards to a loadavg of 200 with 2.6.14-rt5 k1
doesn't seem to have any problems.

I was anxious to get my results "out there" and since -rt5 was still
showing problems I decided to publish with -rt2 anyway (since most of
that work was already done). I made a note of the problems I
experience with these two settings in the report.

Re: latency report

[Trevor Woerner]

On 11/4/05, Trevor Woerner <twoerner.k@gmail.com> wrote:
> I'm trying out -rt6 now.

It just ran out of memory too.