The Function Placement Problem (FPP) Wolfgang Kellerer Technical - - PowerPoint PPT Presentation

▶
the function placement problem fpp
SMART_READER_LITE
LIVE PREVIEW

The Function Placement Problem (FPP) Wolfgang Kellerer Technical - - PowerPoint PPT Presentation

Nov 05, 2022 323 likes •592 views

Chair of Communication Networks Department of Electrical and Computer Engineering Technical University of Munich The Function Placement Problem (FPP) Wolfgang Kellerer Technical University of Munich Dagstuhl, January 16-18, 2017 based on A.

slide-1
SLIDE 1

Chair of Communication Networks Department of Electrical and Computer Engineering Technical University of Munich

Wolfgang Kellerer Technical University of Munich Dagstuhl, January 16-18, 2017

based on A. Basta, W. Kellerer, et al., Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem, ATC’14@ ACM SICGOMM, Chicago, August 2014. and a keynote given at the

  • Intl. Teletraffic Congress, ITC 2016

The Function Placement Problem (FPP)

This work is part of a project that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program grant agreement No 647158 – FlexNets (2015 – 2020).

slide-2
SLIDE 2

NFV: Virtualized network function running in a data center

  • where to place your virtualized network function?
  • what and how to virtualize your function?
  • what are functions‘ interdependencies?

SDN: Control of forwarding path (traverse network functions) and control/data plane split

  • where to place your SDN controllers?

Controller Placement Problem (CPP) (Heller 2012) and a lot of follow up work

  • Controller as a typical network function?
  • no function (de-)composition
  • static placement

2

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM

Function Placement with SDN and NFV

slide-3
SLIDE 3

… not just a generalization of the CPP. Function placement (based on SDN/NFV) needs to consider 1: Function realization: (de-)composition 2: Dynamics: time matters for varying conditions 3: Flexibility: for an overall analysis

3

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM

The Function Placement Problem (FPP)*

* First introduced in A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann, Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem, AllThingsCellular14, Workshop ACM SICGOMM, Chicago, IL, USA, August 2014.

... and many more

slide-4
SLIDE 4

Part 1: Function (de-)composition

4

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM
slide-5
SLIDE 5
  • NFV = ? virtualize & move function (= black box) to DC
  • Consider components/dependencies carefully: function chain

Example: mobile core network functions

5

Part 1: Function Realization à Placement

IP

u-plane MME c-plane PGW HSS PCRF OCS SGW

RAN Core PDN

High volume data traffic High speed packet processing

slide-6
SLIDE 6
  • Virtualization of GW functions [1] à NFV

Function Realization based on NFV

IP

u-plane traffic

Current GW

NE GW-u GW-c data-plane latency? depends on the DC placement network load? traffic transported to DC (longer path à cost)

Virtualized GW

Datacenter

[1] A. Basta et al., A Virtual SDN-enabled EPC Architecture : a case study for S-/P-Gateways functions, SDN4FNS 2013.

slide-7
SLIDE 7

Datacenter

IP

u-plane traffic

  • Decomposition of GW functions [1] via SDN

7

Function Realization based on SDN: move functions back

Virtualized GW

GW-c NE

Decomposed GW

data-plane latency? additional latency is avoided Control load? SDN control load! depends on API (e.g. OpenFlow) GW-u

[1] A. Basta et al., A Virtual SDN-enabled EPC Architecture : a case study for S-/P-Gateways functions, SDN4FNS 2013.

slide-8
SLIDE 8
  • Propagation latency depends on function chain = path SGW - PGW

8

Interdependencies à Function chains (mixed design)

SGW-U PGW-U Datacenter (a) Both SGW and PGW Virtualized u-plane path NE NE SGW-C PGW-C NE+ NE+ SGW-C PGW-C Datacenter SDN API (b) Both SGW and PGW Decomposed CTR SGW-U PGW-C Datacenter SGW-C NE NE+ SDN API (c) SGW Virtualized PGW Decomposed CTR PGW-U PGW-C Datacenter SGW-C NE+ NE SDN API (d) PGW Virtualized SGW Decomposed CTR

Can be more complex for other use cases Function Placement shall address:

  • Function (de-)composition
  • Function chaining
slide-9
SLIDE 9

§Virtualize all GWs? decompose all? mixed deployment?

9

Some Evaluation Studies

  • r

SDN NE

The Functions Placement Problem Part 1

§ minimize core load § satisfy data-plane latency

$

§ Which GWs should be virtualized? decomposed? DC(s) placement?

[2] A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann, Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem, AllThingsCellular14, Workshop ACM SICGOMM, Chicago, IL, USA, August 2014

slide-10
SLIDE 10

less than 4 DCs all virtualized infeasible

§Network load?

10

Evaluation no additional load all decomposed more overhead load overhead vs no. of DCs?

Total Network Load Overhead % SDN Control traffic as % of Data-plane Traffic

normalized by traffic in legacy topology

slide-11
SLIDE 11

Part 2: Dynamic Placement

11

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM
slide-12
SLIDE 12

So far: static placement of functions Reality: requirements (e.g., network traffic) change over time Placement needs to consider

  • change of conditions require to adapt optimal placement à

dynamic (re-)placement

  • migration effort and time
  • Use case:

12

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM

Part 2: Dynamic Placement

Peak dimensioning Tailored dimensioning

Daily network aggregate profile in North America

slide-13
SLIDE 13

11 15 13 12 14 17 18 16 10

SGW PGW

4 3 1 2 9 6 7 8 1 2 3 4 5

Cluster 1 Cluster 2 Cluster3 Cluster 4

population

IP

  • Traffic at each SGW = population * intensity
  • Intensity = f(daytime) [12] and f(time zones)
  • Split day into time slots à change network configuration

Use Case: Traffic Modeling

[12] L. Qian, B. Wu, R. Zhang, W. Zhang, and M. Luo, Characterization of 3G Data-plane Traffic and Application towards Centralized Control and Management for Software Defined Networking," 2013 IEEE International Congress on Big Data

slide-14
SLIDE 14

11 15 13 12 14 17 18 16 10

SGW PGW

4 3 1 2 9 6 7 8 1 2 3 4 5

Cluster 1 Cluster 2 Cluster3 Cluster 4

  • Traffic At each SGW = population * intensity
  • Intensity = f(daytime) [12] and f(time zones)
  • Split day into time slots à change network configuration

Use Case: Traffic Modeling

[12] L. Qian, B. Wu, R. Zhang, W. Zhang, and M. Luo, Characterization of 3G Data-plane Traffic and Application towards Centralized Control and Management for Software Defined Networking," 2013 IEEE International Congress on Big Data

slide-15
SLIDE 15
  • Traffic At each SGW = population * intensity
  • Intensity = f(daytime) [12] and f(time zones)
  • Split day into time slots à change network configuration

Use Case: Traffic Modeling

[12] L. Qian, B. Wu, R. Zhang, W. Zhang, and M. Luo, Characterization of 3G Data-plane Traffic and Application towards Centralized Control and Management for Software Defined Networking," 2013 IEEE International Congress on Big Data

slide-16
SLIDE 16
  • Daily total network load vs. daily DC power saving? à adaptation matters

16

Evaluation 3 DC 2 DC Adaptation

slide-17
SLIDE 17

Part 3: Flexibility as a metric for analysis

17

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM
slide-18
SLIDE 18

Recall: many options to consider for function placement

  • (de-)composition and chaining
  • dynamics

Analyse a network design with respect to the options it can realize to handle dynamically changing requirements: à flexibility as a metric Ex.: Flexibility of a system design w.r.t. function placement

18

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM

Part 3: Flexibility 𝜒"#$%&'&() (𝑒𝑓𝑡𝑗𝑕𝑜. 𝑦) =

∑ ∑ 𝑔𝑓𝑏𝑡𝑗𝑐𝑚𝑓𝑇𝑝𝑚<,>

  • >
  • <

@ 𝑥<,> ∑ ∑ 𝑥<,>

  • >
  • <

all change requests change requests that can be fulfulled by a system design x

slide-19
SLIDE 19

3 design choices to compare for future mobile core network [5]: (1) SDN design (2) NFV design (3) mixed SDN/NFV design Parameter in focus:

  • Flexibility to support different latency requirements for
  • control plane latency and data plane latency

e.g.: {5, 10, 15,…, 45, 50} ms

19

Use Case: EPC Function Placement

[5] W. Kellerer, A. Basta, A. Blenk, Using a Flexibility Measure for Network Design Space Analysis of SDN and NFV, SWFAN’16, IEEE INFOCOM Workshop, April 2016.

slide-20
SLIDE 20

20

Design Choices

(2) NFV design: all functions (data and control) run in a cloud (3) mixed SDN/NFV design: Legacy LTE core design: Gateways (GW) as dedicated middleboxes (1) SDN design: separation of control and data plane for GWs

  • nly control to cloud

control and data to cloud Use Case

slide-21
SLIDE 21

Flexibility measure: Function placement problem formulated as a MILP [6]

  • SDN controllers, mobile VNFs, SDN switches and data centers placement
  • constraints on data and control plane latency
  • weights

21

Flexibility measure and evaluation setup 𝜒"#$%&'&() (𝑒𝑓𝑡𝑗𝑕𝑜. 𝑦) =

∑ ∑ 𝑔𝑓𝑏𝑡𝑗𝑐𝑚𝑓𝑇𝑝𝑚<,>

  • >
  • <

@ 𝑥<,> ∑ ∑ 𝑥<,>

  • >
  • <

𝑥<,> = 𝛽 𝑒𝑏𝑢𝑏𝑀𝑏𝑢𝑓𝑜𝑑𝑧< + 𝛾 𝑑𝑝𝑜𝑢𝑠𝑝𝑚𝑀𝑏𝑢𝑓𝑜𝑑𝑧>

[6] A. Basta, W. Kellerer, M. Hoffmann, H. J. Morper, K. Hoffmann, Applying NFV and SDN to LTE mobile core gateways, the functions placement problem, All things cellular Workshop ACM SIGCOMM, Chicago, August, 2014.

Use Case

slide-22
SLIDE 22

22

Evaluation parameters

Parameters Values Data plane latencies to support {5, 10, 15,…, 45, 50} ms Control plane latencies to support {5, 10, 15,…, 45, 50} ms total: 10 * 10 = 100 possible solutions Data plane latency weight (α) Control plane latency weight (β) α = 1 β = 1 α = 10 β = 1 α = 1 β = 10 Design choices SDN, NFV, SDN/NFV Data center deployment Logically centralized (2 DCs) Distributed (8 DCs) Topology US

Example placement for mixed SDN/NFV design [6]

[6] A. Basta, W. Kellerer, M. Hoffmann, H. J. Morper, K. Hoffmann, Applying NFV and SDN to LTE mobile core gateways, the functions placement problem, All things cellular Workshop ACM SIGCOMM, Chicago, August, 2014.

Use Case

slide-23
SLIDE 23

With respect to the support of latency requirements in function placement:

  • mixed SDN/NFV is more flexible for a logically centralized data center

infrastructure

  • for distributed data centers all three design choices are equally flexible

23

Results [5]

Use Case

[5] W. Kellerer, A. Basta, A. Blenk, Using a Flexibility Measure for Network Design Space Analysis of SDN and NFV, SWFAN’16, IEEE INFOCOM Workshop, April 2016.

slide-24
SLIDE 24

Key Takeaways

  • The Function Placement Problem needs to consider
  • Function (de-)composition
  • Dynamics
  • Flexibility as a new metric for analysis

24

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM
slide-25
SLIDE 25
  • A. Basta, W. Kellerer, M. Hoffmann, H. Morper, K. Hoffmann,

Applying NFV and SDN to LTE Mobile Core Gateways; The Functions Placement Problem, AllThingsCellular14, Workshop ACM SICGOMM, Chicago, IL, USA, August 2014.

  • A. Basta, A. Blenk, M. Hoffmann, H. Morper, K. Hoffmann, W. Kellerer,

SDN and NFV Dynamic Operation of LTE EPC Gateways for Time-varying Traffic Patterns, 6th International Conference on Mobile Networks and Management (MONAMI), Würzburg, Germany, September 2014.

  • W. Kellerer, A. Basta, A. Blenk,

Flexibility of Networks: a new measure for network design space analysis?, arXive report, December 2015.

http://www.lkn.ei.tum.de/forschung/publikationen/dateien/Kellerer2015FlexibilityofNetworks:a.pdf

  • W. Kellerer, A. Basta, A. Blenk,

Using a Flexibility Measure for Network Design Space Analysis of SDN and NFV, Software-Driven Flexible and Agile Networking (SWFAN), IEEE INFOCOM Workshop, San Francisco, USA, April 2016.

25

  • Prof. Wolfgang Kellerer | Chair of Communication Networks | TUM

References for further reading