Re: Processor Scalability and Linux

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



  On 08/09/2010 03:23 PM, Michael Miles wrote:
> JD wrote:
>>     On 08/09/2010 01:37 PM, Michael Miles wrote:
>>
>>> Kwan Lowe wrote:
>>>
>>>> On Sun, Aug 8, 2010 at 1:52 PM, Michael Miles<mmamiga6@xxxxxxxxx>     wrote:
>>>>
>>>>
>>>>
>>>>>>
>>>>> Well, 3D animation is my thing and has been since the Amiga platform.
>>>>> The power to render many minutes of animation and still have functional
>>>>> machine to do the rest of my daily activity.
>>>>>
>>>>> I use a virtual machine running windows 7 for my animation software and
>>>>> if I want to convert a HD movie at the same time as I do everything else
>>>>> it shows a definite slow down.
>>>>>
>>>>>
>>>> I run a 4-node rendering cluster ( dual quad-cores on each, or 32
>>>> cores total and 16G RAM each node).  They're headless and just have
>>>> minimal local disks. All nodes write via bonded 2 x 1Gb Ethernet to a
>>>> fileserver, but network is usually not the bottleneck. When in use,
>>>> CPUs are pegged for hours at a time.  Modeling is done on a quad-core
>>>> Windows 7 system with some relatively high-end ATI cards, but gets
>>>> final render in the cluster. HD conversion is a minor step since the
>>>> renders are done at final resolution.
>>>>
>>>> My point is that it may be more effective to separate your rendering
>>>> hardware. I.e., you can buy a low-end desktop with decent video cards
>>>> that will run your software natively *and* a separate, headless
>>>> compute node that does all the heavy lifting rather than try to bulk
>>>> up a desktop. The desktop will generally have crappy disk i/o, crappy
>>>> memory limits (8G is average), crappy network (wireless or GBit), and
>>>> your CPU will be busy drawing a pretty desktop than actually rendering
>>>> frames.
>>>>
>>>>
>>> --------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>>>
>>>
>>>
>>> I have noticed a bit of a confusing issue.
>>> Lightwave running under Win 7 as a virtual machine under Fedora 12 runs
>>> faster than a native Win 7 machine.
>>> Strange but true.
>>>
>>> It easily shaves off  2 - 3 minutes / frame as a virtual machine.
>>>
>>>
>>> Anyway thanks for the comments.
>>> Question is there a way to have all my cores assigned to one task?
>>> I can easily dedicate the cores to a virtual machine but in a native
>>> Fedora environment I was wondering if I can get all cores to work on one
>>> task.
>>>
>>>
>>> And one other question.
>>>
>>> What software are you using for your render cluster?
>>>
>>>
>>> Way back in the Amiga days I was using Renderman as a rendering farm and
>>> the Screamernet for the Video Toaster.
>>>
>>> I have been doing some experimentation with Blender and it looks very
>>> good but I'm still looking at Lightwave 9 as the best. It is only ported
>>> for Windows though making it a pain as I would like very much to use a
>>> native linux enviroment.
>>>
>>>
>>> It also seems that Lightwave butterfly netrender for linux is here
>>>
>>> http://www.weez.com/2010/08/linux-lightwave-render-farm-getting-bnr-butterfly-netrender-to-work-in-debian-possibly-others/
>>>
>>>
>>>
>>>
>>> We shall see!!!!
>>>
>>>
>>>
>>>
>>>
>> You can - but indirectly.
>> if the process is multithreaded and you want all the cores working on
>> those threads, then
>> when you start up the process:
>> sudo nice -n -10   ProcessPathName
>>
>> will very likely force all threads get on-core before other threads.
>>
>> Danger: There are some system processes that MIGHT get preempted by such
>> a low priority. Se you need to research to see at what priority (nice
>> level) are all the system tasks running.
>>
>>
> That does work but yes, the system had a bird as soon as I pressed enter
> If I wanted to say use 3 out of 4 on a single process and use the 4th
> free core for the system how would I go about that?
>
> Thank you by the way!!!
>
>
> Michael
>
>
To do that, you need a library interface or sysctl command line
that would  "affine" the process and it's threads to
to a set of cpu's (I am not certain if there is granularity here
as far as selecting a subset of cores from a cpu).
I do not know if such lib call or sysctl command
exists for this purpose.
I was aware of it (library call)  in the SVR4-smp kernel.
But it's been a long time and I do not recall the lib call.

-- 
users mailing list
users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines


[Index of Archives]     [Current Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Yosemite News]     [Yosemite Photos]     [KDE Users]     [Fedora Tools]     [Fedora Docs]

  Powered by Linux