Re: MQC class-default

Hi Petr,
your article below is a breakthrough for me in the understanding of
CBWFQ; thank you for that.
I must say that the most misleading information on this subject is the
Cisco doc itself:
http://www.cisco.com/en/US/docs/ios/qos/command/reference/qos_a1.html#wp1011774

"bandwidth-kbps: Amount of bandwidth, in kilobits per second (kbps), to
be assigned to the class. The amount of bandwidth varies according to
the interface and platform in use."

This is just a relative weight! How can the documentation assign a value
in kbps to the class ? :-(

Thanks,
Bit.

On Thu, 2010-04-29 at 14:29 -0700, Petr Lapukhov wrote:
> Hi,
>
> The behavior is different for CBWFQ and HQF to be accurate. CBWFQ is
> merely a WFQ extension and therefore all unclassified traffic is
> serviced using WFQ strategy, unless you assign an explicit weight to
> class-default. Notice that this behavior may seriously starve the
> "class-default" traffic, as all user-defined classes have
> significantly better weights compared to WFQ dynamic weights. You may
> read more at:
>
> http://blog.ine.com/2008/08/17/insights-on-cbwfq/
>
> As for HQF, it uses some round-robin (min-max type) based scheduling,
> which cisco never documented anywhere. By default, all unclassified
> flows are assigned to a single queue with 1% bandwidth reservation
> enforced by the algorithm. This prevents the starvation problem found
> in CBWFQ. Not to mention performance optimizations in HQF compared to
> CBWFQ.
>
> HTH,

Blogs and organic groups at http://www.ccie.net
Received on Fri Apr 30 2010 - 18:18:56 ART