Re: shape on 3560

Marko,
please enjoy the old trees :)

Marko Milivojevic @ 07/08/2011 16:30 -0300 dixit:
> Addition of PQ on Cats completely messes things up for other queues
> and understanding what happens in them.
>
> Even without PQ in the mix, Carlos raises a valid question. Alas, one
> which is impossible to answer.
>
> He's right in saying that shaped queues get a priority in comparison
> to shared. However, this priority is only reflected in the bandwidth
> allocation part. They carve up the total available bandwidth. Whatever
> is left for allocation after shaped queues is carved up by shared
> queues.

That's why in my first post I said (and quote):
   With shaping, the ratio implies a guaranteed rate (you can not
   oversubscribe the port!) and in a sense, it gets prioritized over
   the non shaped queues. The share queues only get to use remaining
   bandwidth.

> Now, when it comes to dequeueing part, this is when things are truly
> interesting. Carlos mentioned a case when the whole port is
> oversubscribed and suggested how the queues will be used. I agree and
> very much disagree with what he wrote. The problem is - it all depends
> on several things a) time interval; b) number of packets in each
> queue; c) size of each individual packet.

Man, I have a problem understanding what you write. That does not mean
that what you write is wrong!
Given that I'm not a native english speaker, would you please let me
know what it means to "agree and very much disagree" ? I'm at loss.

On the other hand, given three CBR feeds @40 Mb each, I would expect
the output to be 40/30/30. You think differently ? That's an easy
test to make too.

> Remember the old CIR = Bc / Tc formula. Well, it still applies. We can
> only send so many bytes in each time interval to satisfy to total
> transmission rate.
>
> Now, what happens when we oversubscribe the port *and* the queues.

If the port is oversubscribed in a consistent manner, queues will
be oversubscribed :)

> Will packet be transmitted if there are available tokens in the
> bucket, but not enough to send the packet? In, for example CQ, the
> answer is yes. in CBWFQ and HQF, the answer is no. How about SRR?
> Cisco doesn't document this (at least I couldn't find it), but this is
> important. Now, what happens if, for example, we run out of bytes in
> an interval to send due to dequeueing method used... Where does the
> next interval start? Q1, or the next queue where the previous interval
> stopped? Again, something Cisco doesn't document, but is rather
> important.

I would expect SRR to use deficit round robin, but I don't see why it
is so important at this point.

> Analyzing QoS is immensely fun topic to me. I think it's an essence of
> networking, but I'll stop now :-). I have some wonderful sequoia trees
> to see!

I completelly agree on the fun part. And we haven't even touched the
incoming priority and 32Gb ring yet ! :)

-Carlos

>
> --
> Marko Milivojevic - CCIE #18427
> Senior Technical Instructor - IPexpert
>
> FREE CCIE training: http://bit.ly/vLecture
>
> Mailto: markom_at_ipexpert.com
> Telephone: +1.810.326.1444
> Web: http://www.ipexpert.com/
>
> On Sun, Aug 7, 2011 at 12:04, Paul Negron <negron.paul_at_gmail.com> wrote:
>> I agree with Marko on this one.
>>
>> The Delay cannot be verified to be 40/30/30 since you cannot measure how
>> many packets from the Priority Q are actually being serviced in relation to
>> the WFQ scheduler that only spits out one packet at a time.
>>
>> Also, what delay are you referencing for a 40/30/30 share. Queuing delay
>> only?
>>
>> What about CPU, Serialization which change pending the packet size. Sure,
>> Serialization delay is almost non-existent with Ethernet designs but it is
>> still there. Thus, you cannot guarantee effective throughput.
>>
>> Paul
>>
>> --
>> Paul Negron
>> CCIE# 14856 CCSI# 22752
>> Senior Technical Instructor
>>
>>
>>
>>> From: gp <gs4me2me_at_gmail.com>
>>> Reply-To: gp <gs4me2me_at_gmail.com>
>>> Date: Sun, 7 Aug 2011 20:16:40 +0200
>>> To: 'Carlos G Mendioroz' <tron_at_huapi.ba.ar>, 'Marko Milivojevic'
>>> <markom_at_ipexpert.com>
>>> Cc: <ccielab_at_groupstudy.com>
>>> Subject: RE: shape on 3560
>>>
>>> So queue which is shaped has guaranteed bandwidth in relation to shared
>>> queues?
>>>
>>> Gp
>>>
>>>
>>> -----Original Message-----
>>> From: Carlos G Mendioroz [mailto:tron_at_huapi.ba.ar]
>>> Sent: Sunday, August 07, 2011 8:05 PM
>>> To: Marko Milivojevic
>>> Cc: gp; <ccielab_at_groupstudy.com>
>>> Subject: Re: shape on 3560
>>>
>>> Marko,
>>> you say that all queues have guaranteed BW but not delay.
>>> I don't fully understand that.
>>>
>>> Are you saying that the delay for a packet of some class is not bounded?
>>> If not, delay is also guaranteed.
>>>
>>> If you have 3 queues, one with 50Mb shape and the other two 1/3 the BW
>>> on a 100 Mb link, anything below congestion (and below 50Mb) will be
>>> just served, but if the 3 queues try to get 40Mb, what will be the
>>> effective throughput of each ?
>>> I would say 40/30/30.
>>>
>>> -Carlos
>>>
>>> Marko Milivojevic @ 07/08/2011 14:22 -0300 dixit:
>>>> All queues are guaranteed bandwidth, but not the delay. SRR is round-robin
>>>> alghoritm. Shaped queues are assigned bandwidth first and whatever remains is
>>>> used by shared queues, but servicing the data in queues is always equal
>>>> between all the queues.
>>>>
>>>> When you enable PQ, Q1 is serviced until empty and only then Q2-4 are
>>>> processed according to the SRR configuration and remaining bandwidth in the
>>>> time interval.
>>>>
>>>> --
>>>> Marko Milivojevic - CCIE #18427
>>>> Senior Technical Instructor - IPexpert
>>>>
>>>> Free CCIE Training: http://bit.ly/vLecture
>>>>
>>>> Mailto: markom_at_ipexpert.com
>>>> Telephone: +1.810.326.1444
>>>> Community: http://www.ipexpert.com/communities
>>>>
>>>> :: Sent from my phone. Apologies for errors and brevity. ::
>>>>
>>>>
>>>> On Aug 7, 2011, at 2:42, "gp" <gs4me2me_at_gmail.com> wrote:
>>>>
>>>>> I read somewhere that shaped queue is processed before shared. Unfortunately
>>>>> cannot find it:)
>>>>>
>>>>> Is it possible to guaranteed some traffic bandwidth to be processed before
>>>>> other with limit in task that it cannot be mapped in priority queue because
>>>>> some other traffic more important is mapped to priority queue?
>>>>>
>>>>> For example:
>>>>> Q1 - traffic X - priority
>>>>> Q2 - traffic XY - less important than traffic X, more important than traffic
>>>>> XYZ
>>>>> Q3&Q4 - traffic XYZ
>>>>>
>>>>>
>>>>> Regards,
>>>>> Gp
>>>>>
>>>>>
>>>>>
>>>>> -----Original Message-----
>>>>> From: Marko Milivojevic [mailto:markom_at_ipexpert.com]
>>>>> Sent: Saturday, August 06, 2011 10:46 PM
>>>>> To: gp
>>>>> Cc: ccielab_at_groupstudy.com
>>>>> Subject: Re: shape on 3560
>>>>>
>>>>> Sharing and shaping has nothing to do with priority queueing. I.e.
>>>>> shaped traffic is not processed before shared - it's only limited to
>>>>> certain percentage of the traffic, while shared queues are simply
>>>>> guaranteed a minimum.
>>>>>
>>>>> You can configure shaping as a percentage of negotiated speed, of
>>>>> course. By default 100% of interface bandwidth (negotiated speed) is
>>>>> available for all queues. If you want to limit Q2 to say, 33% you can
>>>>> configure it as "srr-queue bandwidth shape 0 3 0 0". However, if you
>>>>> want to limit it to say 5 Mb/s REGARDLESS of the negotiated speed,
>>>>> this cannot be done on 3560, as this is a policing and policing is not
>>>>> available in the outbound direction on 3560.
>>>>>
>>>>> --
>>>>> Marko Milivojevic - CCIE #18427
>>>>> Senior Technical Instructor - IPexpert
>>>>>
>>>>> FREE CCIE training: http://bit.ly/vLecture
>>>>>
>>>>> Mailto: markom_at_ipexpert.com
>>>>> Telephone: +1.810.326.1444
>>>>> Web: http://www.ipexpert.com/
>>>>>
>>>>> On Sat, Aug 6, 2011 at 13:19, gp <gs4me2me_at_gmail.com> wrote:
>>>>>> Thank you Marko for explanation, it helps.
>>>>>>
>>>>>> What confusing me if I have a task to guaranteed some traffic for example
>>>>>> 10Mbps on giga interface on switch, and I cannot use priority queue, so I
>>>>>> want to send that traffic before packet in shared queue in way to put it in
>>>>>> shaped queue which will be served before shared queues.
>>>>>>
>>>>>> The problem is that I donb t know what type of interface will be connected
>>>>>> to my switch (1Gbps or 100M) and that is value from which will be
>>>>>> calculated value for shaping on particular queue.
>>>>>>
>>>>>> For example:
>>>>>> Negotiated speed = 1Gbps
>>>>>> Srr-queue bandwidth shape 0 100 0 0
>>>>>> Queue 2 will be shaped to 100Mbps
>>>>>>
>>>>>> Negotiated speed = 100Mbps
>>>>>> Srr-queue bandwidth shape 0 100 0 0
>>>>>> Queue 2 will be shaped to 10Mbps
>>>>>>
>>>>>> In other way as I understand there is no way to hard code shape value
>>>>>> regarding to negotiated speed?
>>>>>>
>>>>>> Regards,
>>>>>> Gp
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Marko Milivojevic [mailto:markom_at_ipexpert.com]
>>>>>> Sent: Saturday, August 06, 2011 7:42 PM
>>>>>> To: gp
>>>>>> Cc: ccielab_at_groupstudy.com
>>>>>> Subject: Re: shape on 3560
>>>>>>
>>>>>> It will be using the bandwidth available to the SRR.
>>>>>>
>>>>>> Cat2#sh mls qos interface fa0/18 queueing
>>>>>> FastEthernet0/18
>>>>>> Egress Priority Queue : disabled
>>>>>> Shaped queue weights (absolute) : 25 0 0 0
>>>>>> Shared queue weights : 25 25 25 25
>>>>>> The port bandwidth limit : 100 (Operational Bandwidth:100.0) <<<<<<<
>>>>>> The port is mapped to qset : 1
>>>>>>
>>>>>> See the "port bandwidth limit" line - that's the BW available to the
>>>>>> SRR. It is derived from the negotiated speed by default, but can be
>>>>>> limited using "srr-queue bandwidth limit" command on the interface.
>>>>>>
>>>>>> Now, let's see if configured bandwidth influences this in any way:
>>>>>>
>>>>>> interface FastEthernet0/18
>>>>>> bandwidth 50000
>>>>>> !
>>>>>>
>>>>>> Cat2(config-if)#do sh mls qos int fa0/18 qu
>>>>>> FastEthernet0/18
>>>>>> Egress Priority Queue : disabled
>>>>>> Shaped queue weights (absolute) : 25 0 0 0
>>>>>> Shared queue weights : 25 25 25 25
>>>>>> The port bandwidth limit : 100 (Operational Bandwidth:100.0)
>>>>>> The port is mapped to qset : 1
>>>>>>
>>>>>> Doesn't look like it does. If I change the bandwidth limit.
>>>>>>
>>>>>> interface FastEthernet0/18
>>>>>> bandwidth 50000
>>>>>> srr-queue bandwidth limit 25
>>>>>> !
>>>>>>
>>>>>> Cat2(config-if)#do sh mls qos int fa0/18 qu
>>>>>> FastEthernet0/18
>>>>>> Egress Priority Queue : disabled
>>>>>> Shaped queue weights (absolute) : 25 0 0 0
>>>>>> Shared queue weights : 25 25 25 25
>>>>>> The port bandwidth limit : 25 (Operational Bandwidth:27.28)
>>>>>> The port is mapped to qset : 1
>>>>>>
>>>>>> Whenever in doubt - ask IOS :-)
>>>>>>
>>>>>> --
>>>>>> Marko Milivojevic - CCIE #18427
>>>>>> Senior Technical Instructor - IPexpert
>>>>>>
>>>>>> FREE CCIE training: http://bit.ly/vLecture
>>>>>>
>>>>>> Mailto: markom_at_ipexpert.com
>>>>>> Telephone: +1.810.326.1444
>>>>>> Web: http://www.ipexpert.com/
>>>>>>
>>>>>> On Sat, Aug 6, 2011 at 01:18, gp <gs4me2me_at_gmail.com> wrote:
>>>>>>> Hello experts,
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> When doing shaping on 3560 interface does reference bandwidth is
>>>>>>> configured
>>>>>>> bandwidth with bandwidth command or negotiated speed?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> For example the switch port is connected on Fast Ethernet router interface
>>>>>>> and I configured bandwidth 1000000 on switch port.
>>>>>>>
>>>>>>> With srr-queue bandwidth shape 0 100 0 0, will queue 2 be shaped on 10 or
>>>>>>> 1
>>>>>>> Mbps?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Thank you!
>>>>>>>
>>>>>>>
>>>>>>> Blogs and organic groups at http://www.ccie.net
>>>>>>>
>>>>>>> _______________________________________________________________________
>>>>>>> Subscription information may be found at:
>>>>>>> http://www.groupstudy.com/list/CCIELab.html
>>>>
>>>> Blogs and organic groups at http://www.ccie.net
>>>>
>>>> _______________________________________________________________________
>>>> Subscription information may be found at:
>>>> http://www.groupstudy.com/list/CCIELab.html
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>> --
>>> Carlos G Mendioroz <tron_at_huapi.ba.ar> LW7 EQI Argentina
>>>
>>>
>>> Blogs and organic groups at http://www.ccie.net
>>>
>>> _______________________________________________________________________
>>> Subscription information may be found at:
>>> http://www.groupstudy.com/list/CCIELab.html
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>
>>

-- 
Carlos G Mendioroz  <tron_at_huapi.ba.ar>  LW7 EQI  Argentina
Blogs and organic groups at http://www.ccie.net