From: Alexei Monastyrnyi (alexeim@orcsoftware.com)
Date: Tue Nov 14 2006 - 15:54:21 ART
I understand all this latency jazz.... With no shaping all happens at
AIR and all is clear.
It is actually interleaving which gives you that low latency,
fragmentation by itself doesn't make any difference.
But I got a feeling that I found an answer on my Q, or at least now all
sounds more logical to me.
My confusion was in how shaper releases traffic. After some off-line
discussion I came to conclusion that it is released as a pointer to next
dequeued (from shaping queue) packet. This is based on knowledge that we
do not rewrite packets from one queue to another but pass pointers to
the packets in buffers. With Bc less than packet size, more than Bc bits
can be released per interval. Shaper though checks if to delay or
release the next packet, according to conforming rate.
In this case CIR doesn't really matter and fragment size should be based
on AIR and desirable Tc/jitter.
My understanding is quite primitive though and may well be wrong, but
now I am more or less is agreement with myself. :-)
Sorry for bothering group with such a minor topic, in the face of
challenges with 4 switches. :-)
Have a good one,
A.
Ivan wrote:
> Fragmentation need to reduce latency delay during serizlization phase packet
> out. In this phase digital representation of packet convert to analog. In
> this case CIR not plays any role.
>
> On Tuesday 14 November 2006 18:41, Alexei Monastyrnyi wrote:
>
>> Just a quick example here - physical interface with AIR 128, one PVC
>> with shaping to CIR 64.
>>
>> map-class frame-relay FRF
>> frame-relay cir 64000
>> frame-relay bc 640
>> frame-relay be 0
>> frame-relay fair-queue
>> frame-relay fragment 80 <- based on CIR
>>
>> vs
>>
>> map-class frame-relay FRF
>> frame-relay cir 64000
>> frame-relay bc 640
>> frame-relay be 0
>> frame-relay fair-queue
>> frame-relay fragment 160 <- based on AIR
>>
>> Why second one is usually said to be recommended?
>>
>> Thanks in advance,
>> A.
>>
>> Alexei Monastyrnyi wrote:
>>
>>> Hi Group.
>>>
>>> I have checked discussions happened to be on this topic starting from
>>> April but haven't found any clear understanding of how one should pick
>>> a fragment size with regards to CIR/AIR of the interface.
>>>
>>> DocCD and Odom's QoS Guide say that one should consider AIR of the
>>> slowest end. In some scenarios CIR (shaping rate) is considered when
>>> picking fragment size.
>>>
>>> Could someone point to pros and cons of either approach?
>>>
>>> TIA,
>>> A.
>>>
>>> _______________________________________________________________________
>>> Subscription information may be found at:
>>> http://www.groupstudy.com/list/CCIELab.html
>>>
>> _______________________________________________________________________
>> Subscription information may be found at:
>> http://www.groupstudy.com/list/CCIELab.html
This archive was generated by hypermail 2.1.4 : Fri Dec 01 2006 - 08:05:47 ART