[Click] Thread Scheduling

[Adam Greenhalgh]

Some of our papers listed here (shameless self plug) will explain some
of the issues associated with core assignments.

http://nrg.cs.ucl.ac.uk/vrouter/

Fairness Issues in Software Virtual Router
Towards high performance virtual routers on commodity hardware

Adam

2009/3/12 Latency Buster <latencybuster at gmail.com>:
> On Wed, Mar 11, 2009 at 11:39 AM, Eddie Kohler <kohler at cs.ucla.edu> wrote:
>> Could you take a look at some handlers?  In particular, what does the
>> "home_thread" handler report for all four PollDevices and ToDevices?  Also
>> the "scheduled" and "tickets" handlers.
>>
>
> Executing click: ./click-uninstall; ./click-install -t 2 --cpu=0 <config>
> And config contains: StaticThreadSched (pd0 0, td0 0, pd1 1, td1 1)
> where pd: PollDevice and td:ToDevice.
>
> While sending bidirectional traffic,
> 1. 'top' returns with only one instant of kclick although pgrep kclick
> provides two processid.
> 2. cat pd0/homethread == pd1/homethread == td0/homethread == td1/homethread = 0
> 3. cat /pd0/tickets ~=  cat /pd1/tickets ~=  cat /td0/tickets ~=  cat
> /td1/tickets
>
> However, I have a "hunch" that the click instance I am running is not
> using all the cores as I am seeing packet drops at 140Mbps (100 bytes)
> while I know that for the same hardware (2xdualcore, opteron 2.6, 8GB
> RAM, 800MhZ bus, Intel 82546EB/pci-x/64bit/100MhZ) MT click can push
> the figures up to 240 Mbps (100 bytes).
>
> Thanks,
> _______________________________________________
> click mailing list
> click at amsterdam.lcs.mit.edu
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>