Network Resource Design - Network Resource Design - Overview Overview ECE/CSC 570: Fall, 2010, Sections 001, 601
Jan 21, 2016
Network Resource Design - Network Resource Design - OverviewOverview
ECE/CSC 570: Fall, 2010, Sections 001, 601
PositioningPositioning Networks must be designed (resource
provisioned) Design should proceed on the basis of
– What use the network is likely to be put to– What behavior is expected or desired from network
Different answers to above questions– Will result in different approach to design
Copyright Rudra Dutta, NCSU, Fall, 2010
Copyright Rudra Dutta, NCSU, Fall, 2010
Network TrafficNetwork Traffic Ultimately, networks exist to serve traffic (enable traffic
to be carried) What is traffic?
– That which occupies / is carried by links Traffic is offered to the network by/at network nodes
– Network is made of end nodes, intermediate nodes, and links– All traffic ultimately originated by end-nodes– However, for hierarchical networks, aggregation may occur
In some network paradigms, E2E traffic is recognizable at all “places” in network
In others, components within aggregated traffic not recognizable inside network
Copyright Rudra Dutta, NCSU, Fall, 2010
Traffic CharacterizationTraffic Characterization Traffic - “Demand” for networking services: b/w and
switching Magnitude (bandwidth)
– Could vary with time, if “reasonably long” life Lifetime
– How long it is resident in the network Arrival and departure patterns
– Call (like telephony) arrival and departure– Increment and decrement– Periodic (scheduled)– Static (long-term)
Requirement of performance– Hard or statistical
Copyright Rudra Dutta, NCSU, Fall, 2010
Network DesignNetwork Design Various aspects of the network must be
determined/chosen/configured Network resources - nodes and links Nodes
– Circuit (physical connection) interface– Buffers, scheduling, routing/forwarding, protocol
Links– Circuit enablement, bandwidth (bitrate capacity), protocol
Goals are in terms of network performance (experienced by traffic)– Basic goal: Connectivity– Basic design methodology: Routing– Others: b/w (if possible), buffer, resource management e.g. link
scheduling– Topology, transmission power, battery allocation, …
Copyright Rudra Dutta, NCSU, Fall, 2010
Issue of Traffic EngineeringIssue of Traffic Engineering Connectivity-only routing (traditional shortest path)
ignores all traffic metrics But traffic exists
– Consider flows 1 4, 1 6, 24, 26
3
4
5
6
44
3
10
1
2
1
5
1
6
Copyright Rudra Dutta, NCSU, Fall, 2010
Network PerformanceNetwork Performance Ultimately, measured in quantities the end-user
cares about– Assuming we have connectivity, now what?
Delay, throughput– Other metrics derived from these
More sophisticated metrics– Predictability of above metrics– Predictability of connectivity: Reliability / Survivability– Predictability of delay or throughput
Guarantees - Quality of Service contracts
– Other emergent characteristics: e.g. Security
Copyright Rudra Dutta, NCSU, Fall, 2010
Designing in Traffic NetworksDesigning in Traffic Networks Controversial proposition:
– “If delay is not important, capacity is not important”– “If delay is important, capacity must be large OR aggregation
must be slotted OR both” Consider the position of router R below
R
1
3
2
4
Q
Copyright Rudra Dutta, NCSU, Fall, 2010
Statistical TDM PerformanceStatistical TDM Performance Bursty traffic, statistical TDM Usual M/M/1 assumptions
– In reality, traffic process is heavier-tailed D(λ, μ) = 1 / (μ - λ) Delay is lower on average: “Statistical Multiplexing Gain”
– But unpredictable for individual packet - prediction is statistical
Link utilization λ/μ
Ave
rag
e D
ela
y (m
s)
R
1
3
2
4
Q
Copyright Rudra Dutta, NCSU, Fall, 2010
M/M/1 QueueM/M/1 Queue
0 1 2 3 4 5 6
p2λ + p4μ = p3 (λ+μ)p1 = p0 -λμ
λ λ λ λ λ λ λ
μ μ μ μ μ μ μ
Network of RoutersNetwork of Routers
Copyright Rudra Dutta, NCSU, Fall, 2010
Copyright Rudra Dutta, NCSU, Fall, 2010
Blocking in TelephonyBlocking in Telephony Delay - very small and constant, operative
quantity is blocking ratio Average call rate λ Average holding time τ Offered traffic load or intensity a = λτ
ac / c! B(a,c) = -------------------
Σk=0 ak / k!c
X Q
Telephone NetworkTelephone Network
Copyright Rudra Dutta, NCSU, Fall, 2010
Copyright Rudra Dutta, NCSU, Fall, 2010
Static Traffic PerformanceStatic Traffic Performance Given “matrix” of traffic demand components
– Static, “always-on”– Usually aggregate– Measured or estimated
Delay - fairly constant for each demand, small Blocking - none; loss - none
– Except in unusual circumstances Performance is measured globally
– Various objectives– Delay or throughput (global, across all components)– Revenue, fairness, protection, …
Copyright Rudra Dutta, NCSU, Fall, 2010
Transport, Demand, CapacityTransport, Demand, Capacity Traffic Networks and Transport Networks Traffic networks: where stochastic demand
picture is operative– Short term switching/routing
Transport networks: where traffic demands of static magnitude are seen to be operative– (Semi-) Permanent– QoS considerations paramount– Demands seen to be injected at transport network
nodes, lower level networks not visible Links must have capacity to carry traffic
– But routing can be designed on basis of traffic
Copyright Rudra Dutta, NCSU, Fall, 2010
Flow Routing and Global RoutingFlow Routing and Global Routing Most general view of routing
– Any part of any flow can be routed along some path from source to destination
Requires the ability to “mark” every part that has to be routed in a distinct manner
– Using labels, or timeslots
3
4
5
6
44
3
10
1
2
1
5
1
6
Copyright Rudra Dutta, NCSU, Fall, 2010
Mathematical Programming ProblemsMathematical Programming Problems A steel company must decide how to allocate production time on a rolling
mill. The mill takes unfinished slabs of steel as input and can produce either of two products: bands and coils. Bands roll off the mill at 200 Tons/hr, generate $25/Ton profit, and at most 6000 Tons per week can be produced. The same figures for coils are 140 Tons/hr, $30/Ton profit, 4000 Tons/wk. If 40 hours of production time are available, what should be produced to maximize profit?
Example adapted from “AMPL”, by Kernighan et al, 1993
Maximize: total profitSubject to: total number of production hours 40
tons of bands produced 6,000tons of coils produced 4,000
Verbal model – Put the objective and constraints
into words– For constraints, use the form
{a verbal description of the LHS} {a relationship} {an RHS constant}
Define the Decision Variables – XB number of tons of bands produced.
– XC number of tons of coils produced.
Construct the Symbolic ModelMaximize:
Subject to:
CB XX 3025 +
( ) ( ) 4014012001 ≤+ CB XX
60000 ≤≤ BX40000 ≤≤ CX
Copyright Rudra Dutta, NCSU, Fall, 2010
Solving LP ProblemsSolving LP Problems
Bands00 2000 4000 6000 8000
Coils
2000
4000
6000Constraints
Feasible region
00 2000 4000 6000 8000
Bands
Coils
2000
4000
6000220K
192K
120K
Profit
Optimal solution
Hours
Graphical Solution Method
Copyright Rudra Dutta, NCSU, Fall, 2010
Solving LP ProblemsSolving LP Problems
Unique Optimal Solution Alternate Optimal Solutions
No Feasible Solution Unbounded Optimal Solution
4 Possible Outcomes
Copyright Rudra Dutta, NCSU, Fall, 2010
Solving LP ProblemsSolving LP Problems Simplex method
– Efficient algorithm to solve LP problems by performing matrix operations on the LP Tableau
– Developed by George Dantzig (1947)– Can be used to solve small LP problems by hand
AMPL and CPLEX– AMPL: modeling language (and software) for designing large and
complex LP/IP problems (now use OPL)– CPLEX: software package (“solver”) to solve large and complex
LP/IP problems Sub-Optimal Algorithms (Heuristics)
– Simulated annealing– Genetic algorithms– Tabu search– Many others, often very specific to the type of problem.
Copyright Rudra Dutta, NCSU, Fall, 2010
Integer ProgrammingInteger Programming
Maximize:
Subject to:
CB XX ′+′ 3000025000
( ) ( ) 4014010002001000 ≤′+′ CB XX
,60 ≤′≤ BX,40 ≤′≤ CX
integer
integer
Convert Example to Integer Program– Assume that orders for bands and coils are placed (and filled) in
1,000s of pounds only.– Although feasible region is greatly reduced, problem becomes much
more difficult. New Symbolic Model
– Let the new decision variables be the number of 1000 pound “units” or orders of bands and coils.
Copyright Rudra Dutta, NCSU, Fall, 2010
Integer ProgrammingInteger Programming
00 2 4 6 8
2
4
6
Feasible integer solutions
Bands
Coils
$185K
Optimal integer solution ($185K)
Graphical Solution Method
Copyright Rudra Dutta, NCSU, Fall, 2010
Multi-Commodity Flow FormulationMulti-Commodity Flow Formulation Parameters
– n : number of nodes– A : set of all links (i, j) – uij : bitrate of link– cij : cost per bit on link– bkl : traffic demand from node k to node l
Variables– xkl
ij : traffic from k to l using link from i to j Goal: minimize total cost
Source: Bertsimas and Tsitsilkis
j
l
i
k
Copyright Rudra Dutta, NCSU, Fall, 2010
Multi-Commodity Flow FormulationMulti-Commodity Flow Formulation
i
Copyright Rudra Dutta, NCSU, Fall, 2010
Management Cycle and DesignManagement Cycle and Design
Near Real-Time
Capacity Mgmt, Netw Engg.
Network Planning
Reactive Protocol Design
Algorithm Design
Resource Design
Copyright Rudra Dutta, NCSU, Fall, 2010
SummationSummation In low level networks, traffic is bursty, unpredictable, and in general
low– A traffic network– Impractical to design for peak traffic, other notions not very meaningful– Design for connectivity, with roughly correct capacities
L3-switched/routed traffic can be thought of as static at a high level of network
– A transport view of network is appropriate, using slotted TDM– This approach is indispensable when strong guarantees must be made
w.r.t. delay, variability of delay, and bandwidth– Capacity of links becomes important in meeting such guarantees– Capacity, routing, and other variables can be thought of as “control
knobs” in the ensuing design problem For circuits, can reflect physical resource occupations to obtain
quantitative idea– May also be useful for “logical” circuits at L3 (or not)