Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
Copyright©2017 NTT corp. All Rights Reserved.P.1
September 28, 2017
APNOMS2017@Seoul
Takashi Miyamura1, Akira Misawa2, Jun-ichi Kani1
1 NTT Laboratories2 Chitose Institute of Science and Technology (CIST)
Design of Optical Aggregation Network with Carrier Edge Functions
Virtualization
2Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1. Background
2. Objective
3. Proposed Method
4. Performance Evaluation
5. Conclude
3Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1.Background2. Objective
3. Proposed Method
4. Performance Evaluation
5. Conclude
Copyright© 2017 NTT corp. All Rights Reserved.
Carrier network architecture
OXC / OADM
IP router
Layer 2 switch
Legend (symbols):
IX
Public cloud(e.g., Amazon)
Internet
OLT
※OLT: Optical Line Terminal, OXC: Optical Cross-connect
Aggregationnetwork
Access net.
Edge router
GW router
OXC
Layer 2 switches
Opt. fiber
Backbone net.
Others
Backbone network:Transport long-haul traffic
Aggregation network:Concentrate user traffic and transport it to the backbone
Access network:Connecting user premises with carrier network
Aggregation network (metro): Transport traffic from access network to backbone network, based on electronic L2 SW.Carrier edge function: provides service-specific functions such as QoS and flow identification at edge router.
Access net.
5Copyright© 2017 NTT corp. All Rights Reserved.
Recent trend in aggregation networks
①: Introducing NFV for Carrier Edge functions
• Edge functions to be relocated from edge routers to distributed pools of commodity servers
②: TDM/WDM-based Optical Aggregation Network
• WDM ring network to accommodate large-volume traffic efficiently
• Improving efficiency by sharing same wavelength through TDM
OLT
OLT
OLT
OLT
Server farms
TDM/WDM ring
Improving efficiency of wavelength through TDM
Accommodate unpredictable traffic demand changes by NFV
VNF: Virtual Network Function
Re-optimize VNF placement according to uneven resource usage
① NFV for edge functions
② TDM/WDM-based network
6Copyright© 2017 NTT corp. All Rights Reserved.
① Carrier edge functions virtualization
Advantages:
• Create VNF for each OLT. The VNF can be flexibly placed on any servers among distributed pools of commodity servers.
• Can be live-migrated from an over-loaded server to an under-utilized one in response to unpredictable traffic changes.
Challenges:
• Determine VNF placement considering traffic demand and geographical location (e.g. distance bet. OLT-Server).
Edge routers(dedicated hardware)
Conventional network
QoS SBC
BAS ACL
NFV
Commodity servers
Edge functions
VNF
#1#2
#3
#4VNF
VNF
VNF
VNF (virtual network function): Implementation as software
7Copyright© 2017 NTT corp. All Rights Reserved.
② TDM/WDM-based aggregations network
Advantages:
• To cope with bursty traffic, enable the same wavelength channel to be shared by multiple OLTs through DBA or other TDM technologies.
Challenges:
• How to efficiently select route for P2MP wavelength path considering bandwidth requirement of each OLT?
[1] K. Hattori et al., IEICE Trans. Electron., Vol. E99-C, No. 2, pp. 189-201,February 2016.
1time 1
timeNE
Optical copler
O/E conversion
1time 12
Access node
WDM link WDM link
2
OLT1
OLT2
ES(Edge server)
DBADBA
DBA
8Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1. Background
2.Objective3. Proposed Method
4. Performance Evaluation
5. Conclude
Copyright© 2017 NTT corp. All Rights Reserved.
Conventional: Deployment of edge routers– Edge routers accommodate fixed group of OLTs based on location. This
causes congested or under-utilized routers due to unpredictable traffic demand changes
Our approach: NFV for carrier edge fucntions– Can accommodate OLTs independent of location, so it can provides
robustness⇒However, consumes more network resources if we allocate VNFs on geographically remote server.
Issues in carrier edge virtualization
Edge servers (ES)
1 2
VNF
OLT
Conventional
Edge routers
3 4 5
Seoul Busan Jeju
Degrade efficiency if traffic demand of OLTs varies widely
1 4 2 3 5
1 2 OLT3 4 5
Improve efficiency to optimize accommodation according to demand changes
Our approach: NFV for edge functions
More resources are consumed if server is far distant from OLT
Seoul Busan Jeju
Copyright© 2017 NTT corp. All Rights Reserved.
• Minimize network cost while efficiently utilizing server resources.
– For optimal VNF placement, we need to consider resource requirements of optical network in addition to server resources
– In optical aggregation network, we need to consider the sharing of wavelength paths as we can use P2MP paths connecting server and OLTs
⇒ Establish optimal design method for VNF placement and P2MP path routing
Objective
1 2 1 2
• Consume 3 WDM links• Disperse server load
• Consume 2 WDM links• Server may be over-loaded
VNF
OLT
ESES
OLT
1
2
1 2
11Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1. Background
2. Objective
3.Proposed Method4. Performance Evaluation
5. Conclude
Copyright© 2017 NTT corp. All Rights Reserved.
We formulate the design problem as mixed integer linear programming (MILP).●Objective function:• Minimize weighted sum of network cost and max server load●Variables:• P2MP wavelength paths between server and OLTs• VNF placement (Select ES to be accommodated)
Proposed design method
Weight parameter
Network cost Max server load
:
Total of WDM links used for P2MP pathsES load
Max load
Copyright© 2017 NTT corp. All Rights Reserved.
Problems1:
Place VNFs on ES
Network model
: ES : OLT
: OADM
Physical topology design
(ES~OLT)
Logical topology design(VNF~OLT)
: VNF
1
2
2
3
1
2
3
1 3
Select ES to accommodate VNF
1 2 3
TDM-WDMnetwork
: Wavelength path
Legends
VNF OLT
Problem 2:
Compute route bet. ES and OLT.
Edge server
OLT
1 2 3
Copyright© 2017 NTT corp. All Rights Reserved.
• VNF placement– Assign binary variables to candidate ES for VNF
placement
• Route for P2MP wavelength path– Configure logical flow bet. VNF~OLT.
– Find route for P2MP path minimize network cost while accommodating traffic demand of logical flows.
Overview of MILP formulations
1 2 3
kVNF
VNF placement P2MP wavelength path
1 2 3 OLT
1 2 3
Traffic
demand
D1
D1+D2
≦Bandwidth of λ
X1k
=1 (accommodate VNF k)
=0 (no placement)
X1k
Binary variable
X2k X3
k
ES
ES
Traffic
demand
D2
15Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1. Background
2. Objective
3. Proposed Method
4.Performance Evaluation5. Conclude
16Copyright© 2017 NTT corp. All Rights Reserved.
Objective of performance evaluation
①: Design of parameter α• Parameter α governs the weight of network cost on server
load
②: Effectiveness• Effectiveness regarding reduction in network cost and
server load
③: Robustness against unpredictable traffic changes
• Evaluate impact of increased variations of traffic demandObjective function
Wight parameter
Network cost Max server load
17Copyright© 2017 NTT corp. All Rights Reserved.
Evaluation conditions
• Topology: 10-node multi-ring
• Traffic demand: Zipf distribution T(x)=b*x-β
• Reference method: MinHop, RoundRobin
Network topology MinHop
RoundRobin
n
1 2 3
OLT
Select server with minimum
hop
1 2 3
i j k
Select server in RoundRobin
orderOLT
OLT
18Copyright© 2017 NTT corp. All Rights Reserved.
0
0.5
1
1.5
2
0 1 10 100 1000
Result①: Design of parameter α
• To configure parameter α around 0~1, we can efficiently disperse server load while avoiding increase in network cost.
Net
wo
rk c
ost
/M
ax s
erv
er
load
Optimization parameter α
Max server load
Network cost
Minimizenetwork cost
Minimizeserver load
19Copyright© 2017 NTT corp. All Rights Reserved.
Result ②-1: Effectiveness
• Compare network cost for various traffic demand
• Proposed method reduced by up to 21%
Net
wo
rk c
ost
Average traffic demand※
(※Normalized by λ bandwidth)
0
0.5
1
1.5
2
0.1 0.2 0.3 0.4 0.5 0.6
Proposed
MinHop
RoundRobin
ReductionBy 21%
10%increase
20Copyright© 2017 NTT corp. All Rights Reserved.
Result ②-2 Effectiveness
• Compare max server load for various traffic demand
• Proposed method reduced by up to 30%
Max
se
rver
load
0
2
4
6
8
0.1 0.2 0.3 0.4 0.5 0.6
Average traffic demand※
(※Normalized by λ bandwidth)
Proposed
MinHop
RoundRobin
ReductionBy 30%
21Copyright© 2017 NTT corp. All Rights Reserved.
0
0.5
1
1.5
2
0.5 1 1.5 2
0
0.5
1
1.5
2
0.5 1 1.5 2
Result ③ Robustness
• Evaluate robustness for various traffic variations β
• Proposed method ensures robustness against traffic variations
Proposed
MinHopRoundRobin
Traffic variations (β)
Proposed
RoundRobin
MinHop
Traffic=50
Net
wo
rk c
ost
Max
se
rve
r lo
ad
Traffic variations (β)
Zipf distribution T(x)=b*x-β
Sharp fluctuations
22Copyright© 2017 NTT corp. All Rights Reserved.
Outline
1. Background
2. Objective
3. Proposed Method
4. Performance Evaluation
5.Conclude
23Copyright© 2017 NTT corp. All Rights Reserved.
Conclusions
• We studied the optimal design of optical aggregation networks with carrier edge function virtualization.• NFV improves efficiency of server resources for carrier
edge functions.
• Consider joint optimization problem that minimizes network cost and maximum server load.
• Formulate the problem as mixed integer linear programming.
• We quantitatively evaluated effectiveness of the optimal aggregation network design with VNF placement.• It reduced the maximum server load and network cost by up
to 30% and 21%, respectively.
• Can improve resource efficiency and robustness to unpredictable demand changes.
24Copyright© 2017 NTT corp. All Rights Reserved.
Thank you
for your kind attention.