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

Re: CBQ vs. TBF



On Mon, 28 Feb 2000 kuznet@ms2.inr.ac.ru wrote:

> Hello!
> 
> > clock (~ .97) to a 2048HZ clock (~ .5ms)?  I don't mind wasting a
> > little CPU to get better control.
> 
> It does not contradict to anything.
> 
> Well, try. I never tried this before.
> 
> Only you also have to fix drivers to shrink their DMA chains
> (it depends on driver, tulip has length 16, which is good for 1msec
> timer) and to add some missing definitions to pkt_sched.h
> (it handles only 100 and 1024 now)
> 
> Alexey


	I will be looking into this.  I will be building a system
shortly that will be perfect to test this on.  As I have to setup a
lab to profile out new Athlon box I will add this to the planned
testing.


	So far on the list I have 

	Testing maxburst values on various flow levels.

	Testing HZ=2048 on packet scheduling.


BTW:  How does this all interact with multiple interfaces?  What
happens when the workload from say 4 100Mbit interfaces doing CBQ is
introduced?  What happens if JIFFIE < workload instead of JIFFIE >
workload?
	

---
As folks might have suspected, not much survives except roaches, 
and they don't carry large enough packets fast enough...
        --About the Internet and nuclear war.