From: Dennis J. Hartmann (dennisjhartmann@hotmail.com)
Date: Fri May 06 2005 - 17:25:52 GMT-3
HOLD ON! We're talking about a lot of concepts here.....
The hold-queue is the entire software queueing system regardless of
whether it's interface or VC based.
The max-threshold is the FIFO queue size for the individual class within
the CBWFQ/LLQ system.
The TX-RING is where the packet goes after it gets scheduled out of the
software queueing system. The TX-RING is FIFO. It can be tuned to allow
more or less packets in this FIFO queue, but this will effect the processor
and it can also have an adverse effect on your QoS scheduling if you make it
too large. FIFO is bad, the WFQ scheduler used in the CBWFQ system is GOOD.
The Priority Queue in LLQ is ALWAYS FIFO though....
The WRED thresholds are very flexible (up to 4,096 packets) because WRED
does not konw how large you have made the queue.
****KEY NOTE****
Remember NOT to compare interface based QoS commands and MQC
"class-based" QoS commands because they're very different. I would
reccomend trying to stick with the MQC style of commands (if possible). The
interface based commands were around first, but a lot of them are being
depracated (rate-limit comes to mind immediately).
I teach the Global Knowledge Cisco QoS class. We don't cover the
interface-based QoS commands in the course.
Sincerely,
Dennis J. Hartmann
White Pine Communications
CCSI#23402/CCIP/CCNP/CCDP/CCNA/CCDA
Cisco IP Voice Support & Design Specialist
Cisco Optical, VPN & IDS Specialist
MCSE
_____
From: sumit.kumar@comcast.net [mailto:sumit.kumar@comcast.net]
Sent: Friday, May 06, 2005 4:17 PM
To: dh8@pobox.com
Cc: ccielab@groupstudy.com
Subject: RE: Shaping vs. Police values
Thanks Dennis,
Since you are a QoS expert here's anothe one for you.
When we use CBWFQ with WRED , the deafult and max queue len in each class is
64. But with WRED we can set the maximum threshold upto 4096. And to top it
we can set vc hold-queue or Frame-relay hold-queue upto 1024 and tx-ring
values too.
So if I visualize this (corect me if I'm wrong) there queues of each class
with length of 64 packets connected to output queue (vc/
interface)connected to TX ring. Where does the WRED thresholds fit in?
thanks
Sumit
-------------- Original message --------------
> I don't "believe" we have to go to that level (figuring out the Tc),
> but I could be wrong. One of the reasons why Cisco is going to smaller Tc
> intervals is because they're pushing to have all of their mechanisms
> optimized for voice over IP. Since the 7900 phones have a default 20ms
> sampling interval, 25ms is close.
>
> You're absolutely right about the Bc and the Be reccomendations,
> although I have seen different numbers in "some" configuration examples
> where the traffic is voice only.
>
> -Dennis Hartmann
>
>
>
> -----Original Message-----
> From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
> sumit.kumar@comcast.net
> Sent: Friday, May 06, 2005 3:48 PM
> To: ccielab@groupstudy.com
> Subject: Shaping vs. Police values
>
> Groupies,
>
> CCIE Prac studies vol-2 : Cisco recommends the Bc vlaue should be equal to
> CIR*1.5 seconds. and Be= Bc*2.
>
> The earlier recommendation for FRTS are Tc= (125ms) To test the router
> defaults I specified a CIR of 2 Mbps w/o Bc and Be with CB-Policing,
> CB-shaping and GTS on interface. Here's what it shows:
>
> For CB-shaping;
> IER7#show policy-map interface
> Traffic Shaping
> Target/Average Byte Sustain Excess Interval Increment
> Rate Limit bits/int bits/int (ms) (bytes)
> 2000000/2000000 12500 50000 50000 25 6250
> Adapt Queue Packets Bytes Packets Bytes Shaping
>
> For CB-Policiing :
> Serial0/0:0
> Service-policy output: test
> cir 2000000 bps, bc 62500 bytes
>
> For GTS on interface:
> #Show run int serial 0/1
> interface Serial0/1
> traffic-shape rate 2000000 50000 50000 1000
>
> Looks like the default Tc for CB-shaping and GTS is set to 25ms but for
> CB-policing is 250 ms Anyone can explain this, what value should we use in
> Lab ?
>
> thanks
> Sumit
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
This archive was generated by hypermail 2.1.4 : Fri Jun 03 2005 - 10:11:56 GMT-3