It's not that it's not supported, it's just that it's not weighted automatically anymore by default. With HQF you do not automatically run WFQ in class-default, you run FIFO. Even if you turn fair-queue on, it's not weighted fair queueing it's flow based fair queueing. This means that the high weighting that RSVP used to get in previous CBWFQ versions doesn't apply anymore. You can still get it to work you just have to manually weight the RSVP traffic however you want in the policy.
For class-default flow based fair queueing is better anyways though, because then you don't have to be as strict with your ingress marking policy. For example if a user is artificially setting all their traffic DSCP EF (e.g their bittorrent traffic), then this will not automatically get a higher weighting in the default class. Instead it will run max-min scheduling against the other flows in the default class, with the final result being a more "fair" distribution of the scheduler.
I really wouldn't worry about something like this in the grand scheme of things though. One of the last remaining applications of RSVP in today's networks is for MPLS Traffic Engineering. Even in that case though, RSVP is not a reservation in the data plane, it's only a reservation in the control plane. This means that if you configure an MPLS TE tunnel with a reserved bandwidth of 100Mbps, the network does not actually enforce this. What I mean by this is that if traffic routing over this tunnel reaches a 1Gbps that is has an output queue that is full, the TE traffic doesn't automatically get 100Mbps of admission to the queue. You'd still need either non-oversubscribed core links or you would need a QoS policy to actually enforce that reservation.
If you *truly* want to see how this works, you would need to setup a test where you have a traffic generator, the router, and then a packet analyzer. You could then generate traffic to the router of various flows and classes, have the router run its outbound queueing, and then see what is received on the traffic analyzer. Based on the outgoing order and timestamps of the packets on packet generator and the receiving order and timestamps on the packet analyzer, you could then deduce how the code of the QoS scheduler is implemented. The problem is that each vendor's QoS implementation is proprietary. It's not like OSPF or BGP where it has to be standardized. Therefore the Cisco documentation and other external information (meaning basically anything other than the source code itself) is not going to give you a truly accurate picture of how the QoS scheduler is actually implemented in IOS.
HTH,
Brian McGahan, CCIE #8593 (R&S/SP/Security)
bmcgahan_at_INE.com
Internetwork Expert, Inc.
http://www.INE.com
________________________________________
From: Carlos G Mendioroz [tron_at_huapi.ba.ar]
Sent: Wednesday, July 04, 2012 7:19 PM
To: Carlos G Mendioroz
Cc: Brian McGahan; dls152_at_cox.net; Cisco certification
Subject: Re: rsvp and HQF
Oh well, this question seems less entertaining than other topics.
-Carlos
Carlos G Mendioroz @ 02/07/2012 06:34 -0300 dixit:
> Brian ?
>
> Carlos G Mendioroz @ 26/06/2012 08:46 -0300 dixit:
>> Brian,
>> since it seems you are the origin of this info, would you please comment
>> on the source ? You just discovered it by testing or there is some
>> documentation that explains why RSVP is not supported with HQF ?
>>
>> -Carlos
>>
>>
>> dls152_at_cox.net @ 25/06/2012 19:55 -0300 dixit:
>>> Carlos,
>>>
>>> Brian states this in this video. I also read that because RSVP is
>>> weighted and HQF is flow fair queueing RSVP whouldn't work.
>>>
>>> http://www.ine.com/all-access-pass/training/playlist/ccie-routing-switching-advanced-technologies-class/fifo-fq-wfq-cbwfq-11212.html
>>>
>>>
>>> ---- Carlos G Mendioroz <tron_at_huapi.ba.ar> wrote:
>>>> Would you please provide a link to some documentation backing that RSVP
>>>> does not work with HQF ?
>>>>
>>>> I can understand that the implementation of intserv RSVP would have to
>>>> be changed in HQF, and that not many people use intserv RSVP, but I
>>>> have
>>>> not seen an official note stating that it does not work.
>>>>
>>>> -Carlos
>>>>
>>>>
>>>> dls152_at_cox.net @ 24/06/2012 18:08 -0300 dixit:
>>>>> Can someone specifically tell me why rsvp doesn't work with HQF? I
>>>>> read that it is because the class default uses flow queueing but
>>>>> what about the created classes? Can you configure class to map
>>>>> protocol rsvp and assign wfq in the policy-map for this class?
>>>>>
>>>>>
>>>>> 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.netReceived on Wed Jul 04 2012 - 20:00:58 ART
This archive was generated by hypermail 2.2.0 : Wed Aug 01 2012 - 15:55:23 ART