1 / 14

Outline brief review ‘you have read the draft’ concepts: dynamic bandwidth reservation & MAR

Max Allocation with Reservation (MAR) BW Constraint Model for MPLS/DiffServ TE & Performance Comparisons ( draft-ash-mpls-dste-bcmodel-max-alloc-resv-01.txt). Outline brief review ‘you have read the draft’ concepts: dynamic bandwidth reservation & MAR analysis of MAR issues conclusions.

kelda
Download Presentation

Outline brief review ‘you have read the draft’ concepts: dynamic bandwidth reservation & MAR

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Max Allocation with Reservation (MAR)BW Constraint Model for MPLS/DiffServ TE & Performance Comparisons(draft-ash-mpls-dste-bcmodel-max-alloc-resv-01.txt) • Outline • brief review • ‘you have read the draft’ • concepts: dynamic bandwidth reservation & MAR • analysis of MAR • issues • conclusions Jerry Ash gash@att.com

  2. MAR Bandwidth Constraint Model • allocates bandwidth to individual class types (CTs) • like the Maximum Allocation Model (MAM) • protects allocated bandwidth by bandwidth reservation methods, as needed, but otherwise fully share bandwidth • meets all requirements for BC models • works well with or without preemption • supports greater efficiency in bandwidth sharing • provides protection of allocated bandwidth under congestion • allows bandwidth sharing in absence of congestion • based on mechanisms in use for 10+ years for multiservice voice/data bandwidth allocation in large-scale networks

  3. Dynamic Bandwidth Reservation • gives preference to certain traffic • for class types (CT) below their BWalloc • on preferred (shortest) path • preferred traffic allowed to seize any idle bandwidth on a link • non-preferred traffic (on CT above BWalloc or on alternate paths) can seize bandwidth only if there is a minimum level of idle bandwidth (called the “bandwidth-reservation threshold”) • on congested link preferred traffic sees low loss while non-preferred traffic sees much higher loss • this situation maintained across wide variation in percentage of preferred traffic • bandwidth reservation robust to traffic variations • as shown in mathematical models & in simulation studies • very widely used in practice

  4. Illustrative Use of MARLink Load States & Allowed Load State • local link states kept of idle link bandwidth • reserved-bandwidth (RBW) • less than RBW + requested BW available • available-bandwidth (ABW) • more than RBW + requested BW available • bandwidth-not-available (BNA) • not enough bandwidth for flow/LSP • allowed load states for flow/LSP setup • when BW < BWalloc any idle link bandwidth can be seized if link not in BNA state • both RBW & ABW states allowed • when BW > BWalloc, links must be in ABW state • RBW state not allowed

  5. Illustrative Use of MAR

  6. Analysis of MAR • options compared • MAR -- flows/LSPs set up with bandwidth reservation • full sharing -- flows/LSPs set up without bandwidth reservation • full-scale 135-switch national network simulation model • 5 CTs -- normal priority voice, high priority voice, normal priority data, high priority data, & best-effort data

  7. Performance Comparison forMAR & Full Sharing Bandwidth Constraint Models6X Focused Overload on Oakbrook(Total Network % Lost/Delayed Traffic)

  8. Performance Comparison forMAR & Full Sharing Bandwidth Constraint Models50% General Overload(Total Network % Lost/Delayed Traffic)

  9. Issues • all BC models MUST meet all requirements • e.g., MUST NOT require use of preemption to work well • many comments on list in support of this • more important now since all BC models are optional • SPs don’t want to get stuck with BC model not meeting requirements • comparisons of BC models • Russian Doll Model (RDM) • can work poorly when preemption not enabled • too much sharing under overload can degrade performance of some CTs • needs to be modified to have acceptable performance when preemption not enabled • MAM & MAR • provide protection of allocated bandwidth under congestion • MAR allows BW sharing in absence of congestion

  10. Issues • protection from pathological traffic patterns & use • issue for all BC models • protect against any possible scenario, however unlikely or atypical? • examples given of bandwidth hogging • could add upper limits on allocated bandwidth to mitigate • but is this necessary? • DS-TE is a bandwidth allocation procedure, involving use of CAC • CAC & bandwidth allocations related to traffic demands • experience doesn’t match bandwidth hogging scenarios • need common assumptions of traffic characteristics & engineering use • assume DS-TE use based on common assumptions

  11. Conclusions • MAR bandwidth constraint model • protects allocated bandwidth by bandwidth reservation methods, as needed, but otherwise fully share bandwidth • meets all requirements for BC models • works well with or without preemption • supports greater efficiency in bandwidth sharing • provides protection of allocated bandwidth under congestion • allows bandwidth sharing in absence of congestion • need common assumptions of traffic characteristics & engineering use • proposed next steps in specifying BC models • specify/progress MAM • specify/progress MAR • hold off on specifying/progressing RDM • needs modification to not perform poorly when preemption not enabled

  12. Backup Slides

  13. Dynamic Bandwidth Reservation Performanceunder 10% Overload

  14. Network Instability Under Congestion • under congestion networks can exhibit “instability” with drastic loss of network throughput • by as much as 50% of traffic carrying capacity • shown mathematically in [NaM73, Kru82, Aki84] & in numerous simulation studies • simple example: fully-connected network with first-choice routing on the 1-link direct path or, if unavailable, on (one of many) 2-link alternate paths • under congestion 1-link direct path often not available & 2-link alternate path may be found and used • 2-link connections take twice the resources as 1-link connections, which leads to more congestion and more alternate routing on 2-link connections • can lead to two possible network states: • most or all connections on 1-link paths (desired condition) • most or all connections on 2-link paths (half the throughput) • solution: use dynamic bandwidth reservation to favor shortest paths vs. longer alternate paths

More Related