Optimal Service Placement using Pseudo Service Chaining Mechanism - - PowerPoint PPT Presentation

• 2016. 11. 15.

Taeheum Na

{taeheum@etri.re.kr}

Network SW Platform Research Section ETRI

Optimal Service Placement using Pseudo Service Chaining Mechanism

IETF 97 meeting @ Seoul

Contents

 Background

• NFV Environment
• Related work

 Pseudo Service Chaining Mechanism

• Phase 1: Calculation of virtual link cost
• Phase 2: Selection of available computing nodes
• Phase 3: Greedy placement

 Conclusion

• 1 -

• 2 -

Background

Our NFV Envir ironment

 Playnet? = Playground of NFV Environment

• Open Source MANO (OSM) based NFV Environment
• Playstore concept
• Extended VIM functionality
• OpenStack (liberty)
• Container and KVM based virtualization
• Using Nova-docker plugin
• Consideration for point-to-point link (E-line type)
• Saving & loading Network Service (NS)
• Save and load NS using VNFFG format
• When NS is loaded, need to consider optimal placement

NFVI (Openstack) VIM (Openstack) Openstack API VIM API VIM Function VIM VNFM NFVO UI/UX VNF EM VNF EM Link Server

• 3 -

Rela lated work

 ETSI Standard

• VNFFG Descriptor (vnffgd)
• Reference type – VLD
• Virtual Link Descriptor (vld)
• Id, vendor, # of endpoints
• Requirement

 Root requirement – BW of E-line, root bandwidth of E-Tree or E-LAN  Leaf requirement – throughput requirement of leaf connection (Tree, LAN)  QoS

• Virtual Link Record (vlr)
• Same Requirement to VLD
• Allocated_capacity – bandwidth allocated for each of the QoS
•  need more specific parameter for the link
• 4 -

Rela lated work

 IETF Standard

• draft-irtf-nfvrg-resource-management-service-chain-03
• 4. Use case

 4.4 Traffic optimization – For efficiency of resource usage, the NFP instances need to be built by default to localize the traffic flows

• draft-lee-sfc-dynamic-instantiation-01
• 3. SFC dynamic instantiation

 Traffic optimization: construct or maintain SFPs to localize the traffic in the network considering load and administrative domains of SFIs and SFLs

•  Our work can be one of the use case in draft document
• 5 -

Rela lated work

 OpenStack – Filter Scheduler

• Step 1: Filtering
• Filtering compute node based on available virtual resources
• Step 2: Weighting
• Ram, I/O operation weight multiplier
• Multiplier can be configured
• Step 3: Sorting
• Largest weighted node have highest priority
• 6 -

• 7 -

Pseudo Service Chaining Mechanism

Pseudo Servic ice Chain inin ing Mechanis ism

 Goal

• By localizing SFs (=Minimize the number of entity in SFPs) based on link description metric
• Saving core network bandwidth
• By avoiding capsulation, save the computation resource
• Getting more better performance of virtual link

 Assumption

• Doesn't consider scaling, failover and policy
• Metric of Link parameter is decided by Operator (SFC user) at first
• Based on monitoring, it can be updated
• 8 -

Pseudo Servic ice Chain inin ing Mechanis ism

 Overview of placement

• Phase 1: Calculation of Virtual Link Costs
• based on VLD parameters calculate link cost
• Selecting pseudo virtual node (PVN)
• Phase 2: selection of available computing nodes
• Phase 3: Placement PVN
• It is recursively conducted
• 9 -

Pseudo Servic ice Chain inin ing Mechanis ism

 Phase 1: Calculation of Virtual Link Costs

• Transaction among service nodes
• Transaction weight at virtual link
• Volume of traffic at virtual link
• 10 -

Table 1. Parameter definitions for calculation of virtual link costs.

Notation Definition

Amount of transactions at a virtual link i Transaction weight for a virtual link i Volume of traffic at a virtual link i Cost of a virtual link i List of virtual links in the order of cost

Pseudo Servic ice Chain inin ing Mechanis ism

 Phase 2: Selection of available computing nodes

• Based on resource requirement of instance
• Available compute node
• 1st available compute node

 Available resource > resource requirement of PVN

• 2nd Available compute node

 Available resource > minimum resource requirement of SN

• Sort in descending order
• 11 -

Pseudo Servic ice Chain inin ing Mechanis ism

 Phase 3: Greedy placement

• Multiple-Knapsack Problem

𝑧𝑗𝑙 1, 𝑗𝑔 𝑞𝑤𝑛𝑙 𝑗𝑡 𝑏𝑡𝑡𝑗𝑕𝑜𝑓𝑒 𝑗𝑜 𝐵𝑂𝑗 0, 𝑝𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 (8)

𝑛𝑏𝑦𝑗𝑛𝑗𝑨𝑓𝑡 𝑨 = 𝑞𝑥𝑙𝑧𝑗𝑙

𝑜 𝑘 =1 𝑛 𝑗=1

(9)

𝑦𝑘𝑙 1, 𝑗𝑔 𝑤𝑛𝑘 𝑗𝑡 𝑏𝑡𝑡𝑗𝑕𝑜𝑓𝑒 𝑗𝑜 𝑞𝑤𝑛𝑙 0, 𝑝𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 (5) 𝑊𝑠

𝑘 𝑦𝑘𝑙 𝑜 𝑘 =1

< 𝑆𝑗 (6)

𝑞𝑥𝑙 = 𝑥

𝑘 𝑦𝑘𝑙 𝑜 𝑘 =1

(7)

• 12 -

Pseudo Servic ice Chain inin ing Mechanis ism

 Phase 3: Greedy placement

• Maximize the sum of cost in the allocated PVM

𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟑 𝑩𝑺𝟑 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟒

𝑺𝟒

𝑻𝑶𝟑

𝑺𝟑

𝑻𝑶𝟓

𝑺𝟓 𝒖𝟒 𝒖𝟑 𝒖𝟐 𝒖𝟑> 𝒖𝟐> 𝒖𝟒

𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟓

𝑺𝟓 𝒖𝟑> 𝒖𝟐> 𝒖𝟒

𝒋𝒈, 𝑩𝑺𝟑> 𝑺𝟑 +𝑺𝟒 𝑩𝑺𝟑< 𝑺𝟐 + 𝑺𝟑 +𝑺𝟒 𝑩𝑺𝟑> 𝑩𝑺𝟐> 𝑩𝑺𝟒 𝑩𝑺𝟑> 𝑩𝑺𝟐> 𝑩𝑺𝟒 𝑸𝑾𝑶𝟐 𝑺𝟑 +𝑺𝟒

𝒏𝒃𝒚( 𝒖𝟐, 𝒖𝟒)= 𝒖𝟐

𝑩𝑶𝟑 𝑩𝑺𝟑−(𝑺𝟑+𝑺𝟒) 𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟑 𝑩𝑺𝟑 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟓

𝑺𝟓 𝒖𝟒 𝒖𝟐 𝒖𝟑> 𝒖𝟐> 𝒖𝟒

𝑸𝑾𝑶𝟐 𝑺𝟑 +𝑺𝟒 𝒋𝒈, 𝑩𝑺𝟑> 𝑺𝟑 +𝑺𝟒 𝑩𝑺𝟑> 𝑩𝑺𝟐> 𝑩𝑺𝟒

Pseudo Servic ice Chain inin ing Mechanis ism

 Recursive operation

• Compare minimum resource requirement
• 14 -

𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟑 𝑩𝑺𝟑 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟒

𝑺𝟒

𝑻𝑶𝟑

𝑺𝟑

𝑻𝑶𝟓

𝑺𝟓 𝒖𝟒 𝒖𝟑 𝒖𝟐 𝒖𝟑> 𝒖𝟐> 𝒖𝟒

𝑩𝑺𝟑> 𝑩𝑺𝟐> 𝑩𝑺𝟒 𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟑 𝑩𝑺𝟑 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟓

𝑺𝟓 𝒖𝟒 𝒖𝟐 𝒖𝟑> 𝒖𝟐> 𝒖𝟒

𝑩𝑺𝟑> 𝑩𝑺𝟐> 𝑩𝑺𝟒

𝑸𝑾𝑶𝟐 𝑺𝟑 +𝑺𝟑 𝑩𝑶𝟐

𝑩𝑺𝟐 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟓

𝑺𝟓 𝒏𝒃𝒚( 𝒖𝟐, 𝒖𝟒)= 𝒖𝟐 𝒖𝟐> 𝒖𝟒

𝑩𝑺𝟐> 𝑩𝑺𝟒> 𝑩𝑺𝟑 −(𝑺𝟑+𝑺𝟒) 𝑩𝑶𝟑 𝑩𝑺𝟑−(𝑺𝟑+𝑺𝟒) 𝒋𝒈, 𝑩𝑺𝟑> (𝑺𝟑+𝑺𝟒) 𝑩𝑶𝟐 𝑩𝑺𝟐 𝑩𝑶𝟒 𝑩𝑺𝟒

𝑻𝑶𝟐

𝑺𝟐

𝑻𝑶𝟓

𝑺𝟓 𝒏𝒃𝒚( 𝒖𝟐, 𝒖𝟒)= 𝒖𝟐 𝒖𝟐> 𝒖𝟒

𝑩𝑺𝟐> 𝑩𝑺𝟒> 𝑩𝑺𝟑 −(𝑺𝟑+𝑺𝟒) 𝑩𝑶𝟑 𝑩𝑺𝟑−(𝑺𝟑+𝑺𝟑)

Conclu lusion

 Result

• Better performance for Loss-rate of UDP
• Decrease round trip time
• Less CPU usage of host node(Interrupt)
• 15 -

Questio ion?

• 16 -

taeheum@etri.re.kr