9761133 劉立恆
DESCRIPTION
Traffic engineering for information-centric networks Martin J. Reed, IEEE ICC 2012 - Next-Generation Networking Symposium. 9761133 劉立恆. Abstract. Information-centric network (ICN) Intra-domain traffic engineering (TE). Information-centric network (ICN). - PowerPoint PPT PresentationTRANSCRIPT
Traffic engineering for information-centric networks
Martin J. Reed, IEEE ICC 2012 - Next-Generation Networking Symposium
9761133 劉立恆
Abstract• Information-centric
network (ICN)
• Intra-domain traffic engineering (TE)
Information-centric network (ICN)
› publish-subscribe Internet routing paradigm (PSIRP)
› Bloom filter (BF)A source routing methodAnother method mentioned is Multi-Protocol Label Switch (MLPS)
› line speed publish-subscribe inter-networking (LIPSIN)LIPSIN = PSIRP + BF
Bloom filter (BF)
› Hash
› False positive
› Forwarding Identifier (FID) in header (routing information)
› Link Identifiers (LID) for each link
Publish-Subscribe Internet Routing Paradigm (PSIRP) with BF = LIne speed Publish-Subscribe Inter-Networking (LIPSIN)
Intra-domain traffic engineering (TE)
› Loss Rate
› Throughput
› MFMC 是 TE 的其中一個目的 (resource reservation)– 降低使用律 ( 最大化沒用到的傳輸能力 )– 應付未來可能增加的需求
• Multicommodity Flow with Minimum Congestion (MFMC)
x(p): path p 的流量 , u(e): edge e 的容量 , ΓL: 剩餘容量d : 需求量
1. 流量小於 edge 的容量 ( 容量已扣除保留的部分 )
2. flow 須滿足需求
• Integer Multicommodity Flow with Minimum Congestion (IMFMC)
對 IP based network ( 同一個 link 不可不同 path)
1. 流量小於 edge 的容量 ( 容量已扣除保留的部分 )
2. flow 須滿足需求每一個 demand (d) 對應一個 path (p)不可將 demand 分割成多條 path
• Maximum Concurrent Flow (MCF)
令 MCFC 等於 MCF 的變體
The Garg and Konemann algorithm
› Lagrange dual problem
› Iterative processing
› ϵ step, ϵ-optimize
› Run time :
Network simulation
› Degree: Weibull distribution ( k=0.42 )
› Edge generation: Viger and Latapylognormal distribution ( μ = 16.6 and σ = 1.04 )
MFMC v.s. Shortest Path Route (SPR)
15 個 nodes20 個虛擬情境
低負載=1
高負載=0
MFMC v.s. IMFMC
高負載=0
低負載=1
15 個 nodes20 個虛擬情境
Variance of IMFMC
高負載=0.1
NPC algorithm
Conclusion
IMFMC MFMC
pros可以使用現有 IP
架構 演算法速度快
cons演算法為 NPC
計算時間差異大無法使用 IP 當
底層架構
延伸討論› Information-Centric
Network 的應用?› IMFMC 的實際架構?› 20 個模擬情境?› MFMC 與理想狀況的
差距?