Virtual Network Resource Allocation Considering Dependability Issues Victor Lira Orientador: Eduardo Tavares
Introduction • Internet notably has a vital role in society; – Entertainment; – Education; – Health; – So on;
Internet’s Ossification Speed, Capacity, New Applications Architecture Innovations (e.g., for better mobility support)
Network Virtualization • Promising approach to deal with Internet’s ossification problem; • Coexistence of multiple instances of virtual networks on a single shared physical infrastructure; • Flexibility in the topology, manageability, scalability and traffic isolation;
Network Virtualization
Dependability • Ability of a system to deliver a particular service in a reliable way; • Metric/attribute of interest: – Availability; • Probability of a system being in a functioning condition. It considers the alternation of operational and nonoperating states;
Proposed Method
PROBLEM FORMULATION
Substrate/Virtual Network • The physical network is represented by an undirected weighted graph G S = (N S , E S ) ; 𝑜 𝑇 ∈ 𝑂 𝑇 → Nodes; 𝑓 𝑇 𝑗, 𝑘 ∈ 𝐹 𝑇 → Links; • A VN request is denoted by 𝐻 𝑊 = (𝑂 𝑊 , 𝐹 𝑊 ) ; 𝐸(𝐻 𝑊 ) → Availability Constraint;
Substrate Network Resources • The remaining or available capacity of a physical node, 𝑆 𝑂 𝑜 𝑇 , 𝑜 𝑇 ∈ 𝑂 𝑇 , is defined by: 𝑆 𝑂 𝑜 𝑇 = 𝑑 𝑜 𝑇 − 𝑑(𝑜 𝑊 ) ∀𝑜 𝑊 ↑𝑜 𝑇 in which 𝑦 ↑ 𝑧 means that the virtual node 𝑦 is mapped on the physical node 𝑧
Substrate Network Resources • Also, the available bandwidth of a path 𝑄 ∈ 𝑄 𝑇 is given by: 𝑓 𝑇 ∈𝑄 𝑆 𝐹 𝑓 𝑇 𝑆 𝐹 𝑄 = 𝑛𝑗𝑜
Virtual Network Allocation • For each VN request received, the VNP accepts or rejects the request, according to the available resources and constraints; • In case of acceptance, a mapping for the VN on the physical network is accomplished, reserving the required network resources;
Virtual Network Allocation • VN mapping is split into activities: (i) node mapping and (ii) link mapping. • Besides, all requests are subject to: 𝐵𝑤 𝐻 𝑊 ≥ 𝐸 𝐻 𝑊
Node Mapping • Each virtual node is mapped into a physical node using 𝑁 𝑂 ∶ 𝑂 𝑊 → 𝑂 𝑇 , so that, ∀𝑜 𝑊 ∈ 𝑂 𝑊 : 𝑑 𝑜 𝑊 ≤ 𝑆 𝑂 𝑁 𝑂 (𝑜 𝑊 )
Node Mapping • If redundancy is adopted, an additional mapping 𝑁 𝑇𝑂 ∶ 𝑂 𝑊 → 𝑂 𝑇 is considered, such that, ∀𝑜 𝑊 ∈ 𝑂 𝑊 , 𝑁 𝑇𝑂 (𝑜 𝑊 ) ≠ 𝑁 𝑂 (𝑜 𝑊 ) subject to: 𝑑 𝑜 𝑊 ≤ 𝑆 𝑂 𝑁 𝑇𝑂 (𝑜 𝑊 )
Node Mapping • In addition, considering cold standby redundancy, ∀𝑜 𝑊 ∈ 𝑂 𝑊 : 𝑁 𝑇𝑂 (𝑜 𝑊 ) ≠ 𝑁 𝑂 (𝑛 𝑊 ) 𝑁 𝑂 𝑜 𝑊 = 𝑁 𝑂 𝑛 𝑊 , 𝑗𝑔𝑔(𝑜 𝑊 = 𝑛 𝑊 ) 𝑁 𝑇𝑂 𝑜 𝑊 = 𝑁 𝑇𝑂 𝑛 𝑊 , 𝑗𝑔𝑔(𝑜 𝑊 = 𝑛 𝑊 )
Link Mapping • The mapping of virtual links to physical paths is defined by 𝑁 𝑁𝐹 ∶ 𝐹 𝑊 → 𝑄 𝑇 (𝑁 𝑂 𝑛 𝑊 , 𝑁 𝑂 𝑜 𝑊 ) , such that, for any 𝑓 𝑊 = (𝑛 𝑊 , 𝑜 𝑊 ) ∈ 𝐹 𝑊 : 𝑆 𝐹 𝑞 ≥ 𝑐 𝑓 𝑊 , ∀𝑞 ∈ 𝑁 𝑁𝐹 𝑓 𝑊
Link Mapping • In VN requests with redundancy, three additional virtual links are required due to redundant nodes: 1. Spare-primary : 𝑁 𝑇𝑄 : 𝐹 𝑊 → 𝑄 𝑇 (𝑁 𝑇𝑂 𝑛 𝑊 , 𝑁 𝑂 𝑜 𝑊 ) ; 2. Primary-spare : 𝑁 𝑄𝑇 ∶ 𝐹 𝑊 → 𝑄 𝑇 (𝑁 𝑂 𝑛 𝑊 , 𝑁 𝑇𝑂 𝑜 𝑊 ) ; 3. Spare-spare: 𝑁 𝑇𝑇 ∶ 𝐹 𝑊 → 𝑄 𝑇 (𝑁 𝑇𝑂 𝑛 𝑊 , 𝑁 𝑇𝑂 𝑜 𝑊 ) ;
Link Mapping • They are mappings from virtual links to physical paths, such that, for any 𝑓 𝑊 = (𝑛 𝑊 , 𝑜 𝑊 ) ∈ 𝐹 𝑊 , 𝑆 𝐹 𝑞 ≥ 𝑐 𝑓 𝑊 , ∀𝑞 ∈ 𝑁 𝑇𝑄 𝑓 𝑊 𝑆 𝐹 𝑞 ≥ 𝑐 𝑓 𝑊 , ∀𝑞 ∈ 𝑁 𝑄𝑇 𝑓 𝑊 𝑆 𝐹 𝑞 ≥ 𝑐 𝑓 𝑊 , ∀𝑞 ∈ 𝑁 𝑇𝑇 𝑓 𝑊
Objective • Allocating VN requests to meet specified constraints (e.g., availability), minimizing the cost resulting from allocations: 𝑓 𝑊 𝑑(𝑜 𝑊 ) 𝑔 + ∗ 𝑦 𝑓 𝑇 𝑓 𝑊 ∈ 𝐹 𝑊 𝑓 𝑇 ∈ 𝐹 𝑇 𝑜 𝑊 ∈ 𝑂 𝑊 𝑓 𝑊 represents the total bandwidth allocated on link 𝑓 𝑇 to the in which 𝑔 𝑓 𝑇 virtual link 𝑓 𝑊 . 𝑦 is an integer variable , which is equal to ‘2’ whenever redundancy is considered on VN request. Otherwise, the value is equal to ‘1’.
DEPENDABILITY MODELING
No redundancy • 𝑁 𝑂 𝑇1 = 𝐵 ; 10 15 A B • 𝑁 𝑂 𝑇2 = 𝐶 ; 25 10 10 11 S1 S2 • 𝑁 𝑁𝐹 𝑇1, 𝑇2 = (𝐵, 𝐶) ; 30 23 VN Request E C D 25 22 12 10 20 Substrate Network Topology
10 15 A B 25 10 10 No redundancy 11 S1 S2 30 23 VN Request C D E 25 22 12 10 20 Substrate Network Topology LINK A-B A B
Hot Standby • 𝑁 𝑂 𝑇1 = 𝐷 ; 10 15 A B • 𝑁 𝑇𝑂 𝑇1 = 𝐹 ; 25 10 10 11 S1 S2 • 𝑁 𝑂 𝑇2 = 𝐸 ; 30 23 VN Request • 𝑁 𝑇𝑂 𝑇2 = 𝐹 ; E C D 25 22 12 10 20 • 𝑁 𝑁𝐹 𝑇1, 𝑇2 = { 𝐷, 𝐸 } ; Substrate Network Topology • ...
10 15 A B 25 10 10 Hot Standby 11 S1 S2 30 23 VN Request C D E 25 22 12 10 20 Substrate Network Topology C LINK C-D D D E LINK D-E D C LINK C-D C LINK D-E E E E E S1-S2 link S1 node S2 node
Cold Standby • 𝑁 𝑂 𝑇1 = 𝐵 ; 10 15 A B • 𝑁 𝑇𝑂 𝑇1 = 𝐶 ; 25 10 10 11 S1 S2 • 𝑁 𝑂 𝑇2 = 𝐷 ; 30 23 VN Request • 𝑁 𝑇𝑂 𝑇2 = 𝐸 ; E C D 25 22 12 10 20 • 𝑁 𝑁𝐹 𝑇1, 𝑇2 = { 𝐵, 𝐷 } ; Substrate Network Topology • ...
10 15 A B 25 10 10 Cold Standby 11 S1 S2 30 23 VN Request C D E 25 22 12 10 20 Substrate Network Topology A_ON C_ON S2 node S1 node A_Repair A_Failure C_Failure C_Repair A_OFF C_OFF Deactivate_B Deactivate_D Activate_B Activate_D B_ON Wait_B D_ON Wait_D B_Failure D_Repair B_Repair D_Failure B_OFF D_OFF LINK_B-D_ON LINK_C_D_ON LINK_A_C_ON LINK_A-C_Repair LINK_A-C_Failure LINK_C_D_Repair LINK_C_D_Failure LINK_B_D_Failure LINK_B_D_Repair S1-S2 link LINK_B_D_OFF LINK_C_D_OFF LINK_A_C_OFF
10 15 A B 25 10 10 Cold Standby 11 S1 S2 30 23 VN Request C D E 25 22 12 10 20 Substrate Network Topology P{( ((#A_ON + #B_ON>0)) S1 node AND ( ((#A_ON > 0)AND(#LINK_A_C_ON > 0)AND(#C_ON > 0)) OR ((#B_ON > 0)AND(#LINK_B_D_ON > 0)AND(#LINK_C_D_ON > 0)AND(#C_ON > 0)) OR ((#A_ON > 0)AND(#LINK_A_C_ON > 0)AND(#LINK_C_D_ON > 0)AND(#D_ON > 0)) OR ((#B_ON > 0)AND(#LINK_B_D_ON > 0)AND(#D_ON > 0)) ) S1-S2 link AND ((#C_ON + #D_ON>0)) S2 node )}
GRASP FOR VIRTUALIZED NETOWRK ALLOCATION
GRASP • GRASP (Greedy Randomized Adaptive Search Procedure); • Two phases: – Construction; – Local search;
GRASP – Construction Phase
GRASP – Local Search
EXPERIMENTAL RESULTS
Experiment Settings • GT-ITM tool to generate the physical network topology; • Substrate network: – 50 nodes randomly conected with probability 0.5; – Nodes capacities and link bandwidths are real numbers uniformly distributed between 50 and 100;
Experiment Settings • 800 VN requests are considered over a period of 50,000 hours; • 0.9 (90%) is the availability constraint for each VN request;
Results - Cost No R edundancy 6000 Hot S tandby C old S tandby 5000 R -ViNE ost 4000 Average C 3000 2000 1000 0 1 9 17 25 33 41 49 Time (thousands of hours)
Results - Availability 1 0,95 Average Availability 0,9 0,85 0,8 No R edundancy 0,75 Hot S tandby C old S tandby R -ViNE 0,7 1 9 17 25 33 41 49 Time (thousands of hours)
Results – Availability ECDF no redundancy 1.0 R-ViNE hot standby cold standby 0.8 P{ X < = x} 0.6 0.4 0.2 0.0 0.80 0.85 0.90 0.95 1.00 Availability
Results – Acceptance Rate 1 0,9 0,8 Average Acceptance Rate 0,7 0,6 0,5 0,4 0,3 0,2 0,1 0 1 9 17 25 33 41 49 Time (thousands of hours) No R edundancy Hot S tandby C old S tandby
Conclusion • Network Virtualization has received particular attention from the scientific community, as several VNs can coexist in the same physical network; • Many algorithms have been proposed to allocate VNs considering performance metrics. However, dependability is usually neglected.
Conclusion • This work proposes a GRASP-based algorithm for allocating virtual networks taking into account dependability issues;
Thanks!
Recommend
More recommend
