Future Internet Chapter 5: Network Function Virtualization 5b: - - PowerPoint PPT Presentation

▶
future internet chapter 5 network function virtualization
SMART_READER_LITE
LIVE PREVIEW

Future Internet Chapter 5: Network Function Virtualization 5b: - - PowerPoint PPT Presentation

Dec 26, 2022 230 likes •941 views

Future Internet Chapter 5: Network Function Virtualization 5b: Foundations & Algorithms Holger Karl Computer Networks Group Universitt Paderborn Goals of this session Previous session has dealt with motivating examples and

slide-1
SLIDE 1

Computer Networks Group Universität Paderborn

Future Internet Chapter 5: Network Function Virtualization 5b: Foundations & Algorithms

Holger Karl

slide-2
SLIDE 2

Goals of this session

  • Previous session has dealt with motivating examples and

architectural fundamentals

  • Open:
  • Which VNF instances to run where?
  • How many instances per VNF are needed?
  • How to interconnect them?
  • This session: What are the relevant algorithmic problems to

solve?

  • The questions, rather than the solutions – no time to go into details
  • f any algorithms
  • Which problems are hard?

2 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-3
SLIDE 3

What to decide in NFV?

  • Given:
  • Topology, annotated with capacities
  • VNF chain to deploy (one or several)
  • Client distribution
  • Geographical or topological
  • Decide:
  • For each VNF in each chain:
  • How many instances to run?
  • Where to execute these instances?
  • How much capacity to assign each instance?
  • For each chain:
  • How to route requests to entry points of the chain?
  • How to route flows between VNFs?

3 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-4
SLIDE 4

Algorithmic sub-problems

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding

4 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-5
SLIDE 5

Simplify the problem

  • Difficulties: Multiple VNFs, interaction of flows, limited

capacities

  • Simplifications
  • Only look at a single VNF, required by many users
  • Idea: Ignore interaction between flows and facilities
  • On edges: latency does not increase for multiple flows, data rate

infinite

  • Facilities have infinite capacity
  • Costs
  • Each facility incurs cost
  • Latency towards facility
  • More generally, path costs
  • Look for Pareto optimal points?
  • Or relate facility and path costs to each other?

5 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-6
SLIDE 6

Excursion: Pareto optimality

  • Suppose a problem has two relevant figures of merit
  • Two metrics
  • Solutions might differ in the degree by which they satisfy

either metric

  • But not possible to trade one metric against the other,

relate into a common, single metric (e.g., “cost”)

  • Then: notion of “dominating solution”
  • Or: Pareto-optimal solutions
  • Set of these: Pareto front

6

Average travel time

  • Avg. # of

traffic casualties

x x x x x x x x x x x x

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-7
SLIDE 7

Overview

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding
  • Testbeds

7 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-8
SLIDE 8

Facility location

  • Goal: Reduce

Round Trip Time

  • Solution idea:

Serve user requests nearby

8

Where to place?

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-9
SLIDE 9

Facility location: Formally

  • Suppose we can weigh facility and path costs
  • Goal: minimize a weighted sum of these costs
  • Given:
  • Graph G=(V,E)
  • Set of possible facility locations F ½ V
  • For each possible location v 2 F, a facility cost f(v)
  • Incurred when opening a facility at that location
  • Demand d(v), 8 v 2 V
  • Path costs: 8 v1, v2 2 V: c(v1, v2)
  • Incurred if demand from location v1 is processed by a facility at v2
  • Usually assumed to be symmetric: c(v1, v2) = c(v2, v1)

9 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-10
SLIDE 10

Facility location: Solution

  • Find
  • a subset F' ½ F of used facilities
  • an assignment of demand to facility: xv,u = 1 if and only if facility at

v serves the demand from u

  • all demand from one location goes to exactly one facility
  • Minimize:
  • 10

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-11
SLIDE 11

Facility location: Main properties

  • NP hard! L
  • In general: Only approximable in O(log |V|) L
  • Better news: If costs are well-behaved, approximation

within 1.861

  • Well behaved: obey triangle inequality
  • I.e., detours never save money: c(u, w) <= c(u, v) + c(v,w)
  • Bad news: The Internet does obey triangle inequality
  • Violations are quite common on BGP graph
  • E.g., https://www.cs.umd.edu/~lume/files/pam09.pdf ,

http://ieeexplore.ieee.org/document/5207998/,

11 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-12
SLIDE 12

Facility location: Greedy algorithm

  • Write the problem as an Integer Programming (IP) problem
  • Solve the relaxed linear version
  • Use this as an initial solution, improve it using a greedy

algorithm

  • Repeat:
  • For all unopened facility, compute the resulting cost when adding it

to the solution

  • Pick the one with best improvement and add it

12

Guha and Kuller, Greedy strikes back: Improved Facility Location Algorithms

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-13
SLIDE 13

Capacitated facility location

  • So far: a facility could serve arbitrary demand
  • Not realistic!
  • Better: Assume a facility has a maximum capacity for the

demand it can handle

  • Next: A local heuristic for capacitated facility location

problem

13

Verter, Uncapacitated and Capacitated Facility Location Problems Keller et al., A local heuristic for latency-optimized distributed cloud deployment

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-14
SLIDE 14

Where to place? – The Model

14

Capacity 1: #Server Capacity 2: #User Server Costs Assignment

Facility Location Problem

+ Opening

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-15
SLIDE 15

Literatur

  • Opt. Problem:

Uncapacitated Capacitated Global:

GreedyFL (1.861-Approx.) No known approximation

Local:

Keller 2013

15 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-16
SLIDE 16

GreedyFL in a Nutshell

  • Every client: Budget
  • Used to pay for
  • 1. Connection cost
  • 2. Facility opening cost
  • Increases every round
  • Client connects if
  • Budget ≥ Connection cost
  • Facility is open
  • Facility opens if
  • Sum of clients’ contribution ≥ facility opening cost
  • Client’s contribution = max(0, budget – distance)

16 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-17
SLIDE 17

GreedyFLL – Sketch

17

Local Algorithm

  • Local participants:

Clients, facilities

  • Coordination messages:

Clients ask facilities

  • Communication limit:

Derived from budget

  • Discrete time model:

Clocked rounds

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-18
SLIDE 18

GreedyFLL – Sketch

18 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-19
SLIDE 19

GreedyFLL – Sketch

19 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-20
SLIDE 20

GreedyFLL – Sketch

20 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-21
SLIDE 21

GreedyFLL – Sketch

21 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-22
SLIDE 22

GreedyFLL – Sketch

22 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-23
SLIDE 23

GreedyFLLC – Sketch

23

Capacity Limit

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-24
SLIDE 24

GreedyFLLC – Sketch

24

Longer runtime, worse RTT

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-25
SLIDE 25

Extensions to facility location

  • Multi-commodity: Different “products” have to be handled

by the facilities

  • Hierarchical: Introduce “backend” facilities that only serve
  • ther facilities, ...
  • E.g., Hub location problem from airlines

25 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-26
SLIDE 26

Overview

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding
  • Testbeds

26 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-27
SLIDE 27

Routing vs. flow problems

  • Typical routing problem: in a graph, find paths that

minimize some cost metric

  • Cost of a path usually a function of the costs of constituting edges

(e.g., sum, max, ...)

  • Often: Costs are independent on what traffic flows along the edges
  • Typical flow problem: Move stream of data along a network

from a source to a sink

  • Where the data stream consumes capacity of the edges
  • Typically allowed to go along multiple paths in parallel

27 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-28
SLIDE 28

Basic single flow problem

  • Given
  • A directed Graph with edge capacities c(u,v) ¸ 0 for each edge (u,v)
  • A source s and a sink t for a flow
  • Solution: a flow f
  • f: V x V ! Real numbers
  • Constraints of a flow
  • Capacity constraint: f(u,v) <= c(u,m)
  • Skew symmetry: f(u,v) = - f(v,u)
  • Flow conversation: å f(u,w) = 0 for all u, except source and sink
  • Skew symmetry eases notation here!
  • Implies: å(u,v) 2 E f(u,v) = å(v,w) 2 E f(v,w) for all v except source or sink
  • Source produces flow, sink consumes it
  • Main point: flow may be split over multiple paths

28 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-29
SLIDE 29

Single flow problem: Example

  • Source: Wikipedia, http://en.wikipedia.org/wiki/Flow_network

29 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-30
SLIDE 30

Single flow problem, some concepts

  • Residual capacity of an edge (with respect to flow f):

cf(u,v) = c(u,v) - f(u,v)

  • Residual network Gf: edge capacities reduced by existing

flow

  • Augmenting path in the residual network: a path from

source to sink, along which all edge capacities are strictly positive

  • Fact: a flow is maximal if and only if there is no augmenting

path

30 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-31
SLIDE 31

Single flow problem: Example

  • Source: Wikipedia, http://en.wikipedia.org/wiki/Flow_network

31 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-32
SLIDE 32

Single flow: Ford Fulkerson algorithm

  • Simple idea:
  • Extend a flow by an augmenting path
  • Can be found by simple breadth-first or depth-first search
  • Repeat until there is no further such path
  • It will always find a maximal flow
  • Complexity: O(VE2)
  • For the Edmonds-Karp variation, which uses breadth-first search
  • Other variants are also all polynomial!

32 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-33
SLIDE 33

Single flow: Variations

  • Goals
  • Maximize flow or minimize capacities
  • Multiple sources or sinks for the same flow
  • Easy to model: introduce super-source, super-sink
  • Connect all sources/sinks with invite capacity edges

33 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-34
SLIDE 34

Multi-commodity flow

  • Natural extension: What happens with multiple flows in a

single network?

  • So-called ``commodities''
  • Each flow competes for the same capacities along the

edges

  • Typical approach: Write it as a linear program
  • Lots of literature on the topic, see e.g., Grag & Konemann
  • Each commodity is a separate chain of VNFs

34

Garg & Konemann, Faster and simpler algorithms for multicommodity flow and other fractional packing problems

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-35
SLIDE 35

Multi-commodity flow: Variations

  • Minimum cost:
  • Edges have costs as well, proportional to used capacity
  • Minimize sum of edge costs
  • Maximum flow:
  • No fixed demands, try to maximize the sum of all flow across

commodities

  • Maximum concurrent flow:
  • Maximize the smallest ratio of a flow compared to the demand of

its commodity

  • Maximum number of satisfied commodities
  • For NFV?

35 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-36
SLIDE 36

Unsplittable flow

  • Flow formulations usually not adequate for practical

networking!

  • Splitting flows arbitrarily across many paths is often not considered

desirable or practical

  • If flows are not allowed to be split (unsplittable flows),

problem becomes much harder

36 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-37
SLIDE 37

Unsplittable flow: Some results

  • Suppose k commodities with source, sink, and demand are

given

  • Goal is to satisfy as many of these commodities as possible
  • Either offline: All commodities are known beforehand
  • Or online: New requests come one by one and have to be either

granted or rejected

  • Much harder problem
  • MaxSNP hard: approximation algorithms exist
  • See e.g. Kolman & Scheideler or Chakrabarti et al.
  • Approximation ratio O(|E|1/2)
  • With matching lower bound !(|E|1/2-²)P
  • Feasible with relatively simple greedy algorithm
  • Improvements exist, but highly depend on graph structure,

structure of commodities and demands, etc.

37 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-38
SLIDE 38

Intermediate summary

  • Suppose we are given a real topology and one/multiple

NFV chains

  • Possible approach so far
  • For each NFV chain, solve a facility location problem to identify

sites where to execute these chains

  • Interconnect them by unsplittable flow computation (separately or

jointly)

  • How about we combined these two steps?

38 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-39
SLIDE 39

Overview

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding
  • Testbeds

39 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-40
SLIDE 40

VNF chain as network to be embedded

  • Alternative thinking: Consider a VNF chain as a

graph/network

  • Consisting of
  • The individual VNFs as nodes
  • Edges to express their data flows
  • Some of the nodes might be tied to specific places in the real

network: sources of client requests

  • Can we map such a (virtual) network into the actual

network?

  • Example of the virtual network embedding problem

40 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-41
SLIDE 41

Virtual network embedding

  • Given two graphs
  • Substrate graph GS = (VS, ES)
  • Virtual graph GV = (VV, EV)
  • Both nodes and edges annotated with capacities (possibly,

different kinds): CPUs, memory, data rate, latency, ...

  • Goal: Find a mapping µ: GV ! GE where
  • A virtual node is mapped to exactly one substrate node
  • A virtual edge is mapped to exactly one substrate path
  • All capacity limits are obeyed

41 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-42
SLIDE 42

Virtual network embedding, variations

  • Relax constraints
  • Map virtual nodes to several substrate nodes
  • But how to coordinate?
  • Map virtual edges to multiple paths
  • In the sense of a flow problem
  • But the underlying network technology must support such multi-path
  • peration
  • Complicate requirements
  • Embed multiple virtual networks at once (offline thinking)
  • Embed virtual networks one after the other, with some networks

leaving again (online thinking)

42 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-43
SLIDE 43

Virtual network embeddeding, solution approaches

  • Option 1: Two separate steps, node mapping and edge

mapping

  • That is what we just talked about in our summary
  • Option 2: One single stage, consider node and edge

mapping in one step

43 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-44
SLIDE 44

Virtual network embedding, a one-step approach

  • Idea: look for isomorphisms between subgraphs of the

virtual network and the actual topology

44

  • Terms and concepts
  • Graph isomorphism in general:

f is an isomorphism between graphs G=(VG, EG) and H =(VH, EH) if and only if for any nodes u, v 2 VG: (u, v) 2 EG $ ( f(u), f(v) ) 2 EH

  • Generalized: ... u, v 2 VG: (u, v)

2 EG $ ( f(u), f(v) ) 2 PH, where PH is the set of paths of graph H

  • Subgraph isomorphism: if G is

isomorphic to some subgraph

  • f H

http://en.wikipedia.org/wiki/ Graph_isomorphism

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-45
SLIDE 45

One-step embedding: Approach

  • Take larger and larger subgraphs of the virtual network
  • Try to embed them
  • If no embedding is found, backtrack, try another subgraph
  • Nice property: Clients' virtual locations can be easily fixed

to real network's actual locations of clients

45

Lischka, Karl: A virtual network mapping algorithm based on subgraph isomorphism detection

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-46
SLIDE 46

One-step embedding: Example

46 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-47
SLIDE 47

Intermediate summary: VNF and network embedding?

  • Facility location: Can decide how many nodes to use to run

VNFs

  • But has no idea of routing
  • Multi-commodity flow:
  • But has no idea of where to place functions
  • Network embedding: Can combine multi-commodity

aspects with placing of nodes

  • But has no idea how to decide on number of decided sites

47 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-48
SLIDE 48

Overview

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding
  • Testbeds

48 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-49
SLIDE 49

Embedding malleable networks

  • We want: a process that combines
  • facility location (decide how many nodes to use) with
  • network embedding (where to place them, how to connect them

with paths)

  • Idea 1: Try to rewrite the VNF chain, add new instances of

nodes, try to embed it, ...

  • Not nice: no idea about actual node capacities, paths, path

capacities, ...

49 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-50
SLIDE 50

Embedding malleable networks: Idea 2

  • Embellish the VNF chain with information about scaling:

how much load can a VNF instance handle, how many

  • ther nodes can it serve?
  • The VNF chain turns into a template, rather than a specific node
  • Use this scaling information in the embedding process; add

new instances if load cannot be handled or path capacities do not allow to reach the nodes

50

Keller, Robbert, Karl: Template Embedding: Using Application Architecture to Allocate Resources in Distributed Clouds

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-51
SLIDE 51

Virtual Network Embedding (VNE)

51

Physical network Virtual network Placement & Embedding = & 1Tb/s 15 Server 200Mb/s FS BS

FS BS

(Application deployment structure)

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-52
SLIDE 52

VNE + demand adaptive scaling

52

Physical network Virtual (application) network

FS BS FS BS FS BS FS BS FS BS FS FS FS BS FS BS FS BS FS

  • Many candidate

virtual networks

  • Embed each of them
  • Feasibility
  • Quality

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-53
SLIDE 53

Idea: Abstract description for a virtual network

53

Physical network Virtual (application) network

FS BS FS BS FS BS FS BS FS BS FS FS FS BS FS FS BS FS BS

  • Many candidate

virtual networks

  • Embed each of them
  • Feasibility
  • Quality

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-54
SLIDE 54

1 3 1 10

Idea: Abstract description for a virtual network

54

Virtual (application) network

FS BS FS BS FS BS FS BS FS BS FS FS FS BS FS FS BS FS BS

Template Graph User FS BS 60 Mb/s 200 Mb/s

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-55
SLIDE 55

VNE + demand adaptive scaling

55

FS BS

Substrate Graph Overlay Graph

FS

èone overlay per demand èadjusted

  • verlay

(network) Here: At most three user for one VM

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-56
SLIDE 56

VNE + demand adaptive scaling

56

FS BS

Substrate Graph Overlay Graph

FS FS

èone overlay per demand èadjusted

  • verlay

(network) Many users è 100s ovlys è 100x VNE Here: At most three user for one VM èmultiple

  • verlays

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-57
SLIDE 57

Idea: Abstract overlay description

57

FS BS FS Template Graph User FS BS 1 3 1 10 Components

& Resource Requirement

Scaling properties Requirement 60 Mb/s describes FS BS FS FS …

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-58
SLIDE 58

Evaluation: Compare embedding approaches

  • Champion: VNE

Challenger: TE

58

O v e r

G e n . C a n d i d a t e s VNE heuristic

infeasible best

Solutions Template Embedding

⌫

One solution

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-59
SLIDE 59

VNE vs TE: Runtime

59

  • VNE
  • Only feasible for small topologies
  • Generating overlays takes a long time

300 1200 2100 3000

  • verlay gen.

vnm)lib

  • Wemp. emb.

ViGeo Game Web 5 8 10 12 5 8 10 12 5 8 10 12 6 12

runWime (V)

templates # user populations >1h

  • gen. VN

VNE TE SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-60
SLIDE 60

VNE vs TE: Success rate

60

In diagram:

  • Only challenging topologies (few resources)
  • Video template

atlanta norway ziE54 ta2 colt cogentco 5 8 1011 5 8 1011 5 8 1011 5 8 1011 5 8 1011 5 8 1011 0.0 0.2 0.4 0.6 0.8 1.0

VucceVV rate TE V1E

(15) (27) (65) (54) (153) (197) # user populations # nodes

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-61
SLIDE 61

Template embedding & flexible chains: Current work

  • Recall previous session: Chains could be specified with

flexible ordering constraints

  • Swap load balancers, firewalls, etc.
  • So far, this does not yet combine scaling abilities
  • Extend this by:
  • Scaling annotations (easy)
  • Optimization problem (ok)
  • Approximations or efficient heuristics (hmmm....)
  • Interactions of multiple chains: Reuse instances?

61 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-62
SLIDE 62

Overview

  • Facility location
  • Routing
  • Splittable vs. unsplittable flows
  • Single vs. multi-commodity flows
  • Network embedding
  • Fixed networks
  • Malleable networks: Template embedding
  • Testbeds

62 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-63
SLIDE 63

Testbeds

  • OpenNFV
  • Not yet available
  • Orchestrators out of different projects (UNIFY, T-Nova,

Alien, ...?)

  • Little known about that outside of their projects
  • Not publicly available
  • Application Deployment Toolkit

63 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-64
SLIDE 64

Application Deployment Toolkit

  • Automates application deployment at multiple sites
  • Application scaling
  • Current load situation, network situation
  • Goal: improve network related performance using smallest amount
  • f resources
  • Modular: exchangeable scaling module
  • Interfaces with

OpenStack, Open Nebular

  • Tests with local testbed

64 http://www.cs.uni-paderborn.de/fachgebiete/fachgebiet-rechnernetze/people/matthias-keller/adt.html

Adaptation Testbed GeoDist Testbed

OpenStack

Cloud Controller

Application Deployment T

  • olkit

Steering-

Controller

Adaptation

Plugin

Resource

Manager

T

  • pology

Manager

Traffic

Manager

TG Switch VM TG VM Host1 Hostn nova comp./

KVM

netem nova comp./

KVM

netem

Control Panel

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-65
SLIDE 65

ADT v2

  • Current plans: Build a new version
  • Better flexibility
  • Pluggable controls
  • SONATA 5GPP project!

65 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-66
SLIDE 66

Scaling up: Maxinet

  • Large-scale NFV setups? With NFV?
  • Tough!
  • Possible starting point: Maxinet
  • Allows to scale up SDN emulations considerably
  • Integrate that with NFV emulations?!

66

https://github.com/MaxiNet/MaxiNet

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-67
SLIDE 67

Currently: ContainerNet

  • Combine Mininet/Maxinet with the ability to execute Docker

containers in virtual topology

  • Provide it with a control interface towards an NFV
  • rchestrator (from SONATA)
  • Allows actual experimentation with reasonable amount of

resources!

  • Currently still under development
  • Contact Manuel Peuster (@UPB) if interested!

SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms 67

slide-68
SLIDE 68

Scenarios?

  • Topologies? Check
  • Topology zoo, SNDLib, ...
  • NFV scenarios?
  • A great big void
  • Very little is known quantitatively about how NFV scenarios look

like

  • Real need in the community - let’s collect!

68 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-69
SLIDE 69

Conclusions

  • Network Function Virtualization is not only an

architecturally challenging area

  • It also abounds with many algorithmic challenges
  • Some well understood, yet still hard
  • Some present new variations of known problems
  • We need to learn to cast aside preconceptions and understand

what is really needed!

  • We also urgently need a reality check!
  • Concrete topologies, concrete NFV chains
  • Concrete numbers! How many chains, load, data rates, how often

new chains, …

  • We need standard benchmarks! We need to grow up as a science!

69 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms

slide-70
SLIDE 70

References

  • See two Mendeley groups:
  • Network Function Virtualization

https://www.mendeley.com/groups/6763981/networkfunctionvirtuali zation/

  • Distributed Cloud Computing

https://www.mendeley.com/groups/6764011/distributedcloudcomp uting/

70 SS 19, v1.1 FI: Ch 5b - NFV Foundations & Algorithms