NTT Network Service Systems Laboratories
Transcript of NTT Network Service Systems Laboratories
NTT Network Service Systems Laboratories
Network on Demand Service with Price-based Resource Control
IEEE/IFIP BoD 2010@OsakaIEEE/IFIP BoD 2010@OsakaIEEE/IFIP BoD 2010@Osaka
T Miyamura K Shiomoto
IEEE/IFIP BoD 2010@Osaka
T Miyamura K ShiomotoT. Miyamura K. ShiomotoNTT Network Service Systems Labs.
T. Miyamura K. ShiomotoNTT Network Service Systems Labs.
NTT Network Service Systems Labs.Today’s BoD service application scenario
• Today’s BoD service– A network provider offers high-capacity bandwidth to its
tcustomers– Basically, the connectivity is limited to Point-to-Point.– Customers often construct a part of their own network byCustomers often construct a part of their own network by
combining multiple point-to-point connections.
Today’s BoD serviceProvider’s network
P2P connection is initiated upon
222Physical network configurationPhysical network configuration
user request
NTT Network Service Systems Labs.Today’s BoD service application scenario cont’d
• Outline of our proposal– The price is basically determined by service parameters (e.g.,
h ldi ti b d idth t )a holding time, bandwidth, etc.)– Customers wish to minimize the total cost while maintaining
adequate service quality.q q y– We consider a new service that offers a minimum-cost
network configuration in combination with multiple point-to-point connectionspoint connections.
Today’s BoD service Our proposal (VNT service)
Point-to-point connection ₉Multi-point connection
33
₉Point-to-point connection
₉Fixed bandwidth
₉Multi point connection
₉Flexible bandwidth & topology
NTT Network Service Systems Labs.Outline of Talk
1. Concept: VNT service2 R t l ith P i i2. Resource control with Pricing3. Numerical Examples3. Numerical Examples4. Concluding Remarks
44
NTT Network Service Systems Labs.Outline of Talk
1. Concept: VNT service2 R t l ith P i i2. Resource control with Pricing3. Numerical Examples3. Numerical Examples4. Concluding Remarks
55
NTT Network Service Systems Labs.
Virtual Network Topology (VNT)
• The network topology formed by optical paths and advertised to the higher layer is called VNT.
• VNT is re configured by optical paths setup/teardown• VNT is re-configured by optical paths setup/teardown.Virtual Network Topology
IP LayerVNT1 VNT2VNT1 VNT2
Optical Layer
6Physical Fiber Topology
NTT Network Service Systems Labs.Our proposal: VNT service
• A service provider network provides continuous connections among multiple customer sites.
– VNT provides a transport among customer’s sitesp p g– Provider accommodates multiple customers (i.e., VNTs)
• Bandwidth of each pipe connecting a pair of customer edges is automatically adjusted in accordance with traffic-demand changesj g
Customer 1 Customer n Traffic demand bet Site A and BCustomer 1 Customer n
A B
h of
w
ork
Residual Bandwidth
Traffic demand bet. Site A and B
VNT1 VNT n
band
wid
ther
’s n
etw
Allocated resources
Tota
l bpr
ovid
eCustomer’s t ffi d d
7Provider’s network
traffic demand
NTT Network Service Systems Labs.Our proposal: VNT service cont’d
• Merits– Customers: improves flexibility regarding bandwidth and p y g g
reduces OPEX and CAPEX– Providers: Improves efficiency of resource utilization
Conventional BoD Service VNT Service
Connectivity Point-to-Point Muti-pointy p
UNI (Data plane) Any (L1/L2/L3) Packet or Frame-based (L2/L3)Connection Triggered by customer Continuously establishedinitiation /release request Continuously established
Allocated bandwidth
Manually specified upon connection request
Automatically adjusted based on actual demand by service bandwidth connection request yprovider
Operation cost for Estimating traffic demand d/ h ldi i i No additional task is required
8
Operation cost for customer network and/or holding time is very
complicated
No additional task is required for traffic demand variation
NTT Network Service Systems Labs.Challenges for VNT service
• Goal (Economics aspect)– We want to maximize the utility of users; at the same time, maximize
the profit of the service providers.p p• Possible problems
i) Each user share the same objective. Resource contention will occur.ii) Network cost increases if resources are provisioned to cover the peakii) Network cost increases if resources are provisioned to cover the peak
demand.• Our approach
W i t d i i h i i j ti ith TE t h l i– We introduce pricing mechanism in conjunction with TE technologies.
Customer 1 Customer ni) Avoid resource contention among users
ii) Reduce network cost by efficient use of resources
⇒ Topology Optimization (TE)
contention among users⇒ Pricing
Provider’s network
NTT Network Service Systems Labs.Outline of Talk
1. Concept: VNT service2 R C t l ith P i i2. Resource Control with Pricing3. Numerical Examples3. Numerical Examples4. Concluding Remarks
1010
NTT Network Service Systems Labs.Our final goal: price-based resource control
• Network resources in a provider's network for the VNT service are shared by multiple VNTs
• Some mechanism is required for avoiding the exhaustion or contention ofSome mechanism is required for avoiding the exhaustion or contention of network resources. – VNT optimization function computes optimal topology for a given demand – Pricing mechanism determines an adequate price and resource allocationPricing mechanism determines an adequate price and resource allocation
considering residual network resources and customer's benefit.
Input VNT Resource
Customer’s traffic demand
VNT optimization Resource allocation
Input Output
esou cemanager (VRM)
Network resource
Customer’s budgetp Resource allocation
PricePricingmechanism
usage
11Provider’s network•Measured from network•Requested by customer
•Feedback
NTT Network Service Systems Labs.Problem we are solving
• Build mathematical model for proposed services and investigate feasibility of price-based resource controlinvestigate feasibility of price based resource control1. Clarify the impact of price on resource allocation and users2 Evaluate effectiveness of optimized topology2. Evaluate effectiveness of optimized topology3. Investigate impact of user behavior (traffic demand
distribution)distribution)
1212
NTT Network Service Systems Labs.Model 1: Overview
• Service provider(SP)’s behavior– Allocate wavelength resources to users
Choose adequate price for a unit of wavelength– Choose adequate price for a unit of wavelength• User’s behavior
– Submit information about peak traffic demand and budgetP t i ti t th t f ll t d– Pay cost in proportion to the amount of allocated resources
• Resource allocation policy– Maximize the benefit of users under given constraints (for simplicity)
Total capacity (C)Total allocated Total capacity (C)
Resource for customer r (xr)
Total allocated resources (∑x )
SP NetworkUser’s Edge
Customer r
Customer 1
Customer 2
OXC WDM link
NTT Network Service Systems Labs.Model 2: Formulation
• Objective– Maximize the sum of user’s
benefit ( utility minus cost)benefit ( utility minus cost)Fig. Utility function
• Determine– Allocated wavelength resources
to each usersw*log(X/Xmin)
to each users– Note that unit price is a given
parameter in our model
• Subject to– Capacity constraint
B d t t i t X– Budget constraint– Service quality constraint
Xmin
* Logarithmic utility function is approximated
14
* Logarithmic utility function is approximated by piece-wise linear function.
NTT Network Service Systems Labs.Outline of Talk
1. Concept: VNT service2 R t l ith P i i2. Resource control with Pricing3. Numerical Examples3. Numerical Examples4. Concluding Remarks
1515
NTT Network Service Systems Labs.Aims of numerical study
Aims:U d t d b i h t i ti f th d– Understand basic characteristics of the proposed architecture and investigate feasibility
Numerical experiments:1. Investigate impact of price1. Investigate impact of price2. Even traffic vs uneven traffic3 Eff ti f ti i d t l3. Effectiveness of optimized topology
1616
NTT Network Service Systems Labs.Conditions: Network and Traffic model
Network model– Simple 3 nodes topology with 32 wavelength per linkp p gy g p
• 3 customers are accommodated in SP network
– Each wavelength is dedicatedly allocated
T ffi d d d lTraffic demand model– Consider stable traffic matrix of even and uneven distribution
• 5 uneven traffic patterns were randomly generated• 5 uneven traffic patterns were randomly generated• Result for uneven traffic is average of 5 patterns
3 Customers Network model Traffic demand model3 Customers Network model Traffic demand model
Even distribution Uneven distribution
32 λ 32 λ
171732 λ
NTT Network Service Systems Labs.Conditions2: Topology optimization
• Simple single-layer optimization is considered. – Optimal route for wavelength path is calculated by our formulation.p g p y
• Optimized topology can utilize entire resources more efficiently.– Fixed topology: Always use a single hop path– Optimized topology: use a detour router in conjunction with sigle hop route
Fixed topology Optimized topology
A
B
A
B
Detour Path
Fiber
A
CPathA
CDirect Path
Detour Path
18
NTT Network Service Systems Labs.Results 1: Impact of Pricing
• Evaluate social welfare and cost while varying price– Fixed topology & Even traffic pattern– Adequate pricing is beneficial for SP and usersq p g
• Avoids congestion and maximize revenue (=cost for users)
Fig. Impact of price on welfare and costg p p
Fig. Resource utilizationModeratelyModerately
usedWelfare used
Under
Welfare
CongestedCongestedUnder-utilized
Cost
1919
NTT Network Service Systems Labs.Results 2: Impact of uneven traffic
• Investigate the traffic patter on social welfare and resource utilization– Uneven traffic leads to inefficient resource utilization and lower welfare– Even traffic outperforms uneven traffic by 20+% for lower price. p y p– High prices suppresses customer’s demand and no significant difference
between even and uneven traffic pattern
Fig. Resource UtilizationFig. Social Welfare
2020
NTT Network Service Systems Labs.Results 3: Fixed vs optimized topology
• Investigate effectiveness of optimized topology under uneven traffic
Compensate uneven resource utilization and improve resource utilization– Compensate uneven resource utilization and improve resource utilization – More effective for lower price (about 10% gain for welfare)– No gain for optimized topology for even traffic distribution (not shown)
Fig. Resource UtilizationFig. Social Welfare
2121
NTT Network Service Systems Labs.Outline of Talk
1. Concept: VNT service2 R t l ith P i i2. Resource control with Pricing3. Numerical Examples3. Numerical Examples4. Concluding Remarks
2222
NTT Network Service Systems Labs.Concluding Remarks
• Conclusion:– The VNT service can improve flexibility regarding bandwidth while reducing
operation overhead and maintaining advantages of existing BoD servicesoperation overhead and maintaining advantages of existing BoD services.– Considering today’s customer’s traffic usage (highly variable and uneven
distribution), topology optimization and resource control with pricing is beneficial.
– It is important to choose adequate price considering traffic demand and residual network resources.residual network resources.
• Further study:– Introduce more sophisticated optimization strategies (e.g., Multilayer TE)– Investigate the merit of service provider (how to utilize effectively residual
bandwidth)bandwidth)– Consider time-varying traffic demand– Establish the pricing strategies
2323
NTT Network Service Systems Labs.Backup Slides 2: Topology seen by user network
• Case 1: User site terminates optical paths routed on the providers network⁺ Some UNI signaling for controlling optical paths is required
• Case 2: User-Provider is packet-based interface. Provider offers IP routing p gfunctionality.⁺ User site regard Provider network as one large router.
BTopology Case 1:
B
Case 2:
A
BSeen by user
A
B
Provider’s network
C
B
C
BVNT
B
A
B
A
2424C
A
C
A
NTT Network Service Systems Labs.Backup Slides 1: Optimizing VNT
• Optical VNT is determined by traffic demand and physical network topology.
Case 1: Demand is high Case 2: Demand is low
BTraffic demand
g
10GbpsB
10Mbps
A
C
10Gbps
10Gbps
A
C1Gbps
600Mbps
VNTB
p
B
p
A A
2525
C C