[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Bandwidth Constraints model



Title: Message
To be more precise, I'd replace third bullet in 2) with following:
 
    "- it can simultaneously achieve isolation, bw efficiency and protection against QoS degradation of the premium CTs. It can simultaneously achieve isolation, bw efficiency but not QoS degradation of lower CTs in presence of the premium/medium classes."
 
Dimitry
 
 
-----Original Message-----
From: Francois Le Faucheur (flefauch) [mailto:flefauch@cisco.com]
Sent: Thursday, October 03, 2002 10:06 AM
To: te-wg@ops.ietf.org
Subject: Bandwidth Constraints model

Hello,
 
Quickly went through the thread we had on BC models before/during Yokohama. My take on it is that:
 
1) There is some concensus on BC Model objectives:
    -i- it should be possible to disable preemption as some network may not want to use preemption
    -ii- when preemption is disabled, the Model should still work "reasonably" well.
    -iii- there should be effective use of bandwidth (ie good bandwidth sharing among CTs).
    -iv- there should be some isolation (ie a CT cannot hogg the bw of another CT).
    -v- there should be protection against QoS degradation (at least of the premium CTs e.g Voice, Premium Data...).
    -vi- the BC model should be reasoanly simple and shouldn't require additional IGP extensions.
 
2) Maximum Allocation Model (MAM) :
    - MAM is "intuitive"/easy to conceptualise
    - In environments where preemption is not to be used, MAM is attractive because:
        * it is good at achieving isolation.
        * if one doesn't worry too much about "QoS degradation" of lower/medium classes, then MAM can achieve reasonable bw efficiency.
    - it cannot simultaneously achieve isolation, bw efficiency and protection against QoS degradation
 
3) Russian Dolls Model (RDM):
    - In environments where preemption can be used, RDM is attractive because it can simultaneously achieve Bw efficiency, isolation and protection against QoS degradation (ie -iii-, -iv- and -v-)
    - in environments where preemption is not to be used, RDM cannot achieve good isolation.
 
4) Other Models:
    - Jerry indicated it is actually possible to achieve -iii-, -iv-, -v- without preemption and will document another BC Model (maximum allocation with reservation (MAR)) which achieves this.
 
Are we in synch?
 
Cheers
 
Francois