Finding the Optimal Reconfiguration for Network Function Virtualization Orchestration with Time-varied Workload Satyajit Padhy, Jerry Chou National Tsing Hua University The Third International Workshop on Systems and Network Telemetry and Analytics (SNTA2020)
Outline 1. Background 2. VNF Placement Problem a. Approaches 3. VNF Reconfiguration Problem a. Problem Definition 4. Objective & Approach 5. Results 2
Introducing NFV (Softwarizing Middleboxes) DDoS VNF ● Less CAPEX/OPEX ● More Flexibility Traditional Middleboxes ● No-Fixed Location Server ● Introduce new services to the network seamlessly ● Better resource and energy utilization ● For the Future (5G, SDN) 3 ** NFV=Network Function Virtualization, VNF= Virtual Network Function
Outline 1. Background 2. VNF Placement Problem a. Approaches 3. VNF Reconfiguration Problem a. Problem Definition 4. Objective & Approach 5. Results 4
VNF Placement SFC Service Function Chain (SFC) is group of ● DDoS Firewall QoE monitor Protection chained Network Functions (NF). Ordering required NFs are processed in virtualized instances ● Virtual resources called VNFs. Firewall DDos QoE Protection Goal is to place VNFs in physical network. Monitor ● The NFs must be processed in the same ● Virtualization Layer order as SFC Sever Sever 5 Physical resources
Static VNF Placement Approach SFC1 NF1 NF2 NF3 ● Objective: Minimize energy cost (4) (4) (4) SFC2 ● Input: Resource demands of SFC NF1 NF2 NF3 (4) (4) (2) ● Output: VNF Placement SFC3 ● Constraint: Resource constraint NF4 NF2 NF3 (4) (4) (2) R_demand(VNFs) < R_capacity(PM) ○ ● Approach: Fit as many VNFs in a VNF3(8) PM so that less PMs are used VNF2(12) VNF1(8) VNF4(4) 6 PM 3 (25) PM 1 (25) PM 2 (25)
Time-Varied Workload Challenges SFC1 NF1 NF2 NF3 (8) (6) (6) Workload demands usually change ● SFC2 over time. NF1 NF2 NF3 (6) (4) (2) Resource violation can occur. ● SFC3 NF4 NF2 NF3 How to process the increased ● (6) (4) (2) demands? Energy cost can still be reduced ● VNF3(8) Solution: VNF Reconfiguration ● VNF2(12) VNF1(8) VNF4(4) 7 PM 3 (25) PM 1 (25) PM 2 (25)
Outline 1. Background 2. VNF Placement Problem a. Approaches 3. VNF Reconfiguration a. Solutions b. Problem Definition 4. Objective & Approach 5. Results 8
VNF Reconfiguration SFC1 NF1 NF2 NF3 Scale-up (8) (6) (6) SFC2 Increase the resources of a VNF ● NF1 NF2 NF3 Constraint: There should be ● (6) (4) (2) enough residual capacity in PMs SFC3 NF4 NF2 NF3 (6) (4) (2) Scale up VNF4( 6 ) VNF2( 14 ) VNF3( 10 ) VNF1(8) PM 3 (25) PM 1 (25) PM 2 (25) 9
VNF Reconfiguration SFC1 NF1 NF2 NF3 Scale-up (8) (6) (6) Increase the resources of a VNF ● SFC2 Constraint: There should be enough residual ● NF1 NF2 NF3 capacity in PMs (6) (4) (2) SFC3 Scale-out NF4 NF2 NF3 (6) (4) (2) Scale out Create a new instance of a VNF ● Constraint: Redirection of selected ● SFCs can cause delay VNF4( 6 ) VNF3( 10 ) VNF2( 14 ) VNF1( 6 ) VNF1( 8 ) 10 PM 3 (25) PM 1 (25) PM 2 (25)
VNF Reconfiguration Scale-up SFC1 Increase the resources of a VNF ● NF1 NF2 NF3 Constraint: There should be enough residual ● (8) (6) (6) capacity in PMs SFC2 NF1 NF2 NF3 Scale-out (6) (4) (2) SFC3 Create a new instance of a VNF ● NF4 NF2 NF3 Constraint: Redirection of selected SFCs VNF ● (6) (4) (2) Migration can cause delay Migration VNF4( 4 ) VNF3( 10 ) VNF2( 14 ) Migrate VNF to another PM ● Constraint: Migrating all SFCs can VNF1( 14 ) ● cause delay, downtime etc. PM 3 (25) PM 1 (25) PM 2 (25) 11
Problem Definition: VNF Reconfiguration Problem Each SFC s with a chain of NFs {n 1 , n 2 ,...n f } INPUT OUTPUT SFCs NF <--> VNF Placement Resource demand VNF resource capacity Bandwidth requirement VNF V jk , j is type of VNF and k is instance INPUT OUTPUT VNFs Resource Capacity VNF <--> PM Placement PM Resource capacity Reconfiguration Solution PM status PM P i , i th PM Multi-tenant VNF 12 Type of NF = Type of VNF
Outline 1. Background 2. VNF Placement Problem a. Approaches 3. VNF Reconfiguration Problem a. Problem Definition 4. Objective & Approach 5. Results 13
Objectives & Approach ● The objective is to address the tradeoff between reducing the energy cost and VNF reconfiguration. ○ Which reconfiguration solution should be chosen? ● Approach: Integer Linear Programming (ILP) formulation to get an optimal solution 14
ILP Formulation 15
ILP Formulation: R 11 R 21 R 31 Resource Constraints 4 6 6 SFC 1 1. Constraint 1 * VNF Resource Constraint n1 n2 n3 Type(n) == Type(j) VNF 31 VNF 11 VNF 21 2. Constraint 2 * PM Resource capacity constraint * VNF <--> PM M 1 M 2 16 25 35 P 1 P 2
ILP Formulation: Placement Constraints SFC 1 3. Constraint 3 * VNF Placement constraint * VNF should be placed on active PM only n1 n2 n3 VNF 31 4. Constraint 4 VNF 11 VNF 21 * Each VNF should be placed on one PM only A 1 1 A 1 2 17 1 1 P 1 P 2
ILP Formulation: Link Constraints SFC 1 5. Constraint 5 * Link Capacity Constraint n1 n2 n3 VNF 31 6. Constraint 6 VNF 11 VNF 21 * Flow conservation constraint SFC1 18 P 1 P 2
ILP Formulation: X t-1 211 X t-1 221 X t-1 231 Objective Function 1 1 1 Minimize Energy Cost + Migration Cost + Instantaniation Cost n1 n2 n3 VNF 31 VNF 11 VNF 21 VNF 12 19 P 1 P 2
Outline 1. Background 2. VNF Placement Problem a. Approaches 3. VNF Reconfiguration Problem a. Problem Definition 4. Objective & Approach 5. Results 22
Experimental Setup ● Service Function Chains Number of SFCs: 60 ○ Number of NFs in one SFC: 5 ○ Resource demands randomly generated ○ ● Physical Infrastructure: ○ Number of PMs: 50 ○ Heterogeneous resources ○ Bandwidth capacity of physical link: 1Gbps ● Solver: IBM CPLEX Network services consisting of Network Address Translation ( NAT ), Firewall ( FW ), Traffic Monitor ( TM ), WAN Optimization Controller ( WOC ), Intrusion Detection System ( IDC ) and Video 23 Optimization Controller ( VOC )
Results: Reconfiguration Solutions Reconfiguration solutions with ● different traffic arrival rate Initially scale up is the most ● preferred solution As traffic increases, residual ● capacity of PM decreases Scale out becomes more preferred ● than Migration as traffic rate increases 24
Prioritize Migration Cost Prioritize Energy Cost Prioritize Instantaniation Cost 25 Results: Tuning ILP Parameters
Results: Importance of Reconfiguration Costs Only Migration has the highest cost ● since it’s oblivious Migration + Scale out helps in ● reducing cost further All three methods gives the least ● cost since ILP solver can use both scale out/up to reduce reconfiguration cost further. 26
Conclusions ● Reconfiguration solutions are highly beneficial for time-varied workload. ● There can be a tradeoff between reducing energy cost and VNF reconfiguration cost. ● We have proposed a ILP formulation for VNF reconfiguration problem. ● We have used a two level placement solution to solve this problem. ● Preferred reconfiguration solution: ○ Scale up is the most preferred ○ As traffic increases, scale out is preferred more ● As work in progress, we will propose a heuristic solution for this problem. 27
Th Thank yo nk you Please don’t hesitate to reach us if you have any questions 28
Recommend
More recommend
