[Click] Click running on multi-core processor

[Beyers Cronje]

Hi,

You need to use StaticThreadSched
http://read.cs.ucla.edu/click/elements/staticthreadsched element to tell
Click how to allocate tasks to threads. Without this element Click will use
only one thread.

A simple example on how to use is:

fd0::FromDevice(eth0) -> Queue -> td1::ToDevice(eth1);
fd1::FromDevice(eth1) -> Queue -> td0::ToDevice(eth0);

StaticThreadSched(fd0 0, td1 0, fd1 1, tf0 1);

This config schedules fd0 and td1 on thread 0 and the other two elements on
thread 1. It is highly recommended to read the following publication
http://pdos.csail.mit.edu/papers/click:usenix01/usenix01.pdf which delves
into much more details around multithreading and Click and how to best use
it.

Also, highly recommended to upgrade to latest version of Click, there have
been many multithreading stability and performance changes implemented
inside Click since v1.8

Also note not all elements are multithread safe.

Beyers


On Fri, Dec 2, 2011 at 1:08 AM, Bingyang LIU <bjornliu at gmail.com> wrote:

> Hi all,
>
> I'm using click for performance evaluation of IPSec. But it seems that only
> one core of the processor is used. I'm wondering how Click can fully use
> the multi-core for computation to achieve high performance.
>
> The detail of my experiment is as follows:
>
> * Hardware: CPU Inter Xeon X3210 (quad core at 2.13Ghz), 4GB RAM. (a server
> on deterlab)
> * Software: Ubuntu8.04 + Click1.8, with PollDevice enabled.
> * Configuration: ./configure --with-linux=/usr/src/linux-2.6.24.7
> --enable-ipsec --enable-warp9 --enable-multithread=4 (I also tried 16, but
> the outcome is similar with 4)
> * mode: kernel
>
> The outcome:
> * The click router has the loss rate of about 0.5% when incoming packet
> rate is only 200000 pps (packet size = 64B). If I remove the ipsec modules,
> the loss rate is about 0.1%.
> * I checked the CPU usage. Anyway, the click used only one cpu.
> * If there is no input packet, the cpu usage is as follows, I guess
> PollDevice consumes the cpu time:
> Cpu0  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
> Cpu1  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
> Cpu2  :  0.0%us, 48.0%sy,  0.0%ni, 25.0%id,  0.0%wa,  0.0%hi, 27.0%si,
>  0.0%st
> Cpu3  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
> * If there is some input packets, the cpu usage is as follows. It seems
> that the IPSec module only consumes a small portion of the CPU time, which
> causes low performance.
> Cpu0  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
> Cpu1  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
> Cpu2  :  0.0%us, 68.8%sy,  0.0%ni, 25.0%id,  0.0%wa,  0.0%hi,  6.2%si,
>  0.0%st
> Cpu3  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,
>  0.0%st
>
> So what I'm wondering is that, is there any solution to my problem?
> Or is there any easy way to fully use the CPU resource, or achieve
> parallelism in a multi-core router? Is the paper "Flexible Control of
> Parallelism in a Multiprocessor PC Router" the only available solution?
>
> thanks!
> best
> Bingyang
> --
> Bingyang Liu
> Network Architecture Lab, Network Center,Tsinghua Univ.
> Beijing, China
> Home Page: http://netarchlab.tsinghua.edu.cn/~liuby
> _______________________________________________
> click mailing list
> click at amsterdam.lcs.mit.edu
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>