The Wheezy PREEMPT_RT kernel works well on the newer hardware I've
tested on.  Plenty good enough if you've got "hardware assist", and
maybe OK for with software step-gen if you don't need super high rates.
 RTAI/Xenomai jitter numbers are still about 2-4x better, but both are
significantly more invasive patchsets, while the PREEMPT_RT code is
gradually getting mainlined.

Details on use and some performance numbers are on the wiki:

http://wiki.linuxcnc.org/cgi-bin/wiki.pl?Debian_Wheezy_Linux-Rt_Compile_LinuxCNC

On 2/21/2013 1:34 AM, Lars Segerlund wrote:
> On the a somewhat related note, debian wheezy have RT kernels
> included, just for information I haven't tried it out yet. But it's
> nice to have prebuilt.
> 
> 2013/2/20 Michael Haberler <[email protected]>:
>>
>> Am 20.02.2013 um 22:07 schrieb Tom Easterday:
>>
>>> Let me re-ask the question, for which I gave too much prior information. 
>>> Apologies for that.  Are there pre-built packages for RTAI on Ubuntu 12.04 
>>> Precise?
>>> -Tom
>>
>> no, which was part of the reason for the Xenomai/RT-PREEMPT effort
>>
>> but you will very likely be able to run precise with the existing RTAI 
>> kernel just fine by manual installation with dpkg
>>
>> - Michael
>>
>>
>>>
>>>
>>>
>>> On Feb 20, 2013, at 3:52 PM, Michael Haberler <[email protected]> wrote:
>>>
>>>> Tom,
>>>>
>>>> Am 20.02.2013 um 20:29 schrieb Tom Easterday:
>>>>
>>>>> I have a problem that looks like either process scheduling, or buffering, 
>>>>> some sort of blocking on a linuxcnc host running 12.04/Xenomai and 
>>>>> 2.6.0pre.  It does not display this problem on 10.04/RTAI and 2.6.0pre 
>>>>> (same base hardware, an intel atom board).   The symptom is that when 
>>>>> continuously jogging, a DRO showing position on a remote device changes 
>>>>> smoothly when talking to 10.04/rtai.  On 12.04/xenomai we get a distinct 
>>>>> pattern of three updates and a pause, three updates and a pause.  The 
>>>>> pause is some number of milliseconds long (1/4 second maybe) and 
>>>>> distinctly visible.  The remote connection is through our web server 
>>>>> (Rockhopper) over tcp (websocket) on wired ethernet.
>>>>>
>>>>> We first thought it was packet buffering or network latency and played 
>>>>> with TCP parameters and system buffers but that had no effect.  We put 
>>>>> timing statements in the web server code at the point were we write the 
>>>>> data out and it appears that the whole web server process is being 
>>>>> blocked or something.  We tried re-nicing the web server but it didn't 
>>>>> help.  If we run Axis locally while doing this it doesn't display this 
>>>>> pause, it moves along smoothly as one would expect.
>>>>>
>>>>> For debugging I thought I might try 12.04 and rtai to see if it was just 
>>>>> with the xenomai kernel or if rtai on 12 does it as well.  Are there rtai 
>>>>> packages available for 12.04 yet?  I was going through the email list 
>>>>> archive on the new RT stuff but don't see any pointer to packages on rtai 
>>>>> specifically...
>>>>>
>>>>> Any thoughts on what we might look at to narrow it down further?
>>>>
>>>>> From the above I can tell:
>>>>
>>>> - you are running some of your own code because you cite some 'web server' 
>>>> which is not part of a released or working branch. Where is the code, the 
>>>> configuration you are using, and the exact sequence of commands leading up 
>>>> to it?
>>>> - you say you are running '2.6.0pre', whatever that is, that tag isnt a 
>>>> unique identifier. Please state repository, branch name, commit, and which 
>>>> changes you did.
>>>> - you do not say anything about hardware, exact lan setup including any 
>>>> intermediate hardware, the board, kernel version - how are we supposed to 
>>>> guess?
>>>>
>>>> With that level of information, my thought is that you will get zero help 
>>>> in narrowing down things further, simply because it is impossible.
>>>>
>>>> --
>>>>
>>>> that said, what I suggest you do is: triage the problem into a) network 
>>>> driver and hardware b) tcp kernel stack and throughput c) application.
>>>>
>>>> a) Find throughput tests which measure at the packet loss level with the 
>>>> least kernel intervention, that is - raw throughput tests at the ethernet 
>>>> or UDP level. Find out if there are significant differences between 
>>>> operating systems on the same hardware. Change hubs/switches or whatever 
>>>> have in between. Try a crossover cable.
>>>>
>>>> b) do the same including the TCP stack which measures end-to-end 
>>>> throughput at the TCP level. Observe the TCP retransmission numbers in 
>>>> netstat which should give a clue.
>>>>
>>>> c) as for application level issues, install wireshark and pcap, and trace 
>>>> the flow between the programs which show the the issue - we dont know 
>>>> which they are since you do not tell. The packet capture will include 
>>>> timestamps and TCP retransmissions which hint at network problems. If you 
>>>> see any retransmissions, try alternate hardware and drivers - this zoo has 
>>>> many animals. Try with everyting on one machine to exclude network issues.
>>>>
>>>> And: read this: http://www.chiark.greenend.org.uk/~sgtatham/bugs.html
>>>>
>>>> --
>>>>
>>>> One reason why I suggest you produce a more thorough report, and not some 
>>>> ephemeral problem narrative: I spent several hours recently in a off-list 
>>>> debugging session on a 'reported Xenomai problem'. The problem was finally 
>>>> resolved by me demanding to 'yank that ethernet hub *now* and throw it in 
>>>> the wastebin' in no uncertain words. No more Xenomai issues since. Btw 
>>>> that was a device 'which always used to work so far'. So much for 
>>>> perceived causality.
>>>>
>>>>
>>>> - Micahel
>>>>
>>>>
>>>>
>>>>
>>>> ------------------------------------------------------------------------------
>>>> Everyone hates slow websites. So do we.
>>>> Make your web apps faster with AppDynamics
>>>> Download AppDynamics Lite for free today:
>>>> http://p.sf.net/sfu/appdyn_d2d_feb
>>>> _______________________________________________
>>>> Emc-developers mailing list
>>>> [email protected]
>>>> https://lists.sourceforge.net/lists/listinfo/emc-developers
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Everyone hates slow websites. So do we.
>>> Make your web apps faster with AppDynamics
>>> Download AppDynamics Lite for free today:
>>> http://p.sf.net/sfu/appdyn_d2d_feb
>>> _______________________________________________
>>> Emc-developers mailing list
>>> [email protected]
>>> https://lists.sourceforge.net/lists/listinfo/emc-developers
>>
>>
>> ------------------------------------------------------------------------------
>> Everyone hates slow websites. So do we.
>> Make your web apps faster with AppDynamics
>> Download AppDynamics Lite for free today:
>> http://p.sf.net/sfu/appdyn_d2d_feb
>> _______________________________________________
>> Emc-developers mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/emc-developers
> 
> ------------------------------------------------------------------------------
> Everyone hates slow websites. So do we.
> Make your web apps faster with AppDynamics
> Download AppDynamics Lite for free today:
> http://p.sf.net/sfu/appdyn_d2d_feb
> _______________________________________________
> Emc-developers mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/emc-developers
> 


-- 
Charles Steinkuehler
[email protected]

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_feb
_______________________________________________
Emc-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-developers

Reply via email to