[Click] Re: about ToDevice

[Eddie Kohler]

Yihong,

Tasks run very frequently -- much less than 1/100s granularity.  Click Timers 
don't have inherent 1/100s granularity either; they work differently than you 
are thinking.  The only reason you'd need to worry is if you have some CPU-bound 
process that will take its entire timeslice (until the next clock interrupt). 
The correct answer is to raise the kernel's HZ value.

Eddie


Yihong Zhou wrote:
> Eddie :
> 
> Thanks for your reply. As I understand, the resolution for click timer is
> about 0.01 seconds.  I don't know how often the task scheduling list would
> be polled.  Since we need to keep the timing for 802.11 packets processing,
> we may need a timer or task scheduler that has the granularity of
> microseconds.  Would that be possible in click?  Or we may need to implement
> a timer with smaller granularity?
> 
> ----- Original Message ----- 
> From: "Eddie Kohler" <kohler at cs.ucla.edu>
> To: "Yihong Zhou" <zyihong at ece.utexas.edu>
> Cc: <click at amsterdam.lcs.mit.edu>
> Sent: Wednesday, June 23, 2004 1:11 AM
> Subject: Re: about ToDevice
> 
> 
> 
>>Hi Yihong,
>>
>>Well, generally speaking, the ToDevice will fall off the "task scheduling
> 
> list"w
> 
>>when the queue is empty, and go onto the "task scheduling list" when the
> 
> queue
> 
>>becomes nonempty.  This is assuming that the ToDevice is fed by exactly
> 
> one
> 
>>Queue.  The "task scheduling list" consists of tasks that are run at a
> 
> very high
> 
>>frequency, polling-style.
>>
>>And no, you don't need to patch the kernel to run click user-level.
>>
>>Eddie
>>
>>
>>Yihong Zhou wrote:
>>
>>>Hi, Eddie:
>>>
>>>I have a simple question:
>>>
>>>How is "ToDevice" scheduled?  Let's say, it pulls packets from the queue
>>>so that the click can control the timing of packet processing. ToDevice
>>>is derived from AnyTaskDevice, and there is _Task private data member.
>>>Is it scheduled periodically, or is it triggered by some other events?
>>>
>>>By the way, if we want to run click in the user level, do we need to
>>>patch the linux kernel?
>>>
>>>Thanks a lot.
>>>
>>>
>>