Software-Defined Networks Mayutan Arumaithurai , Jiachen Chen , Edo - - PowerPoint PPT Presentation

software defined networks
SMART_READER_LITE
LIVE PREVIEW

Software-Defined Networks Mayutan Arumaithurai , Jiachen Chen , Edo - - PowerPoint PPT Presentation

Exploiting ICN for Flexible Management of Software-Defined Networks Mayutan Arumaithurai , Jiachen Chen , Edo Monticelli , Xiaoming Fu and K. K. Ramakrishnan * University of Goettingen, Germany * University of California,


slide-1
SLIDE 1

Exploiting ICN for Flexible Management of Software-Defined Networks

Mayutan Arumaithurai†, Jiachen Chen†, Edo Monticelli†, Xiaoming Fu† and K. K. Ramakrishnan*

† University of Goettingen, Germany * University of California, Riverside, U.S.A.

slide-2
SLIDE 2

Middleboxes usually provide these services for reasons of

– Policy control, security, performance optimization – They have to be resident on the path of a flow =>

  • traffic might have to deviate from its natural “IP” shortest path

and forced through middleboxes

Problem Statement and Terminology

9/26/2014 Exploiting ICN for Flexible Management of SDN 2

Objective of this paper: Deal with an important and common problem, i.e. Service Chaining Long term vision: Enhance SDN with Information Centricity to improve network management Service Chaining: The steering of flows through the different network functions/services needed, before it is delivered to the destination.

slide-3
SLIDE 3

Base Topology

9/26/2014 Exploiting ICN for Flexible Management of SDN 3 Middlebox Prefix Popper ICN-Switch Ingress Egress Firewall A DSA R5 Cache DPI DPI Firewall B

slide-4
SLIDE 4
  • Middleboxes performing additional processing of packets

before it is delivered has become an integral need of the Internet

– E.g. , Facebook, Twitter, FourSquare, Google Instant, MyYahoo require

content to go through middleboxes in order to improve performance

– Dynamic site accelerators – TCP optimization – NAT – Proxy – Firewall – DPI – CDNs

Motivation - I

4 9/26/2014 Exploiting ICN for Flexible Management of SDN

slide-5
SLIDE 5
  • Advent of Network Function Virtualization (NFV)

– makes it easier to deploy middleboxes in a flexible and dynamic

manner

– can be instantiated, removed and positioned on demand

NFV based Middleboxes

5 9/26/2014 Exploiting ICN for Flexible Management of SDN

Managing such an NFV based middlebox provisioning and service chaining can be challenging

slide-6
SLIDE 6
  • Software Defined Networking (SDN’s)

– attempt to perform such network management by making use of a

logically centralized controller

  • Controller has a global view of the network and therefore network mgmt.

is easier than in the case of the current distributed approach

– Setup flow based forwarding rules on paths – Provides greater control for networks to steer packets flexibly without

being constrained by traditional routing such as OSPF, BGP

SDN - I

6 9/26/2014 Exploiting ICN for Flexible Management of SDN

However, the management logic (“what” is required) is intrinsically coupled with the node location (“where” it is available).

slide-7
SLIDE 7
  • Intrinsic coupling results in issues pertaining to =>

– Flexibility: cannot make real time decisions since location is decided – Scalability: places flow based forwarding rules on every router – Reliability: when middleboxes fail, a new path has to be built

  • or a backup path for every flow and every possibility (i.e. if router A fails, if

router B fails and so on) exist

SDN - II

7 9/26/2014 Exploiting ICN for Flexible Management of SDN

However, the management logic (“what” is required) is intrinsically coupled with the node location (“where” it is available).

slide-8
SLIDE 8

We argue that the performance of SDN can be further improved by using ideas of information centricity

– Decouple location from the identity (name) of the function – Can make better use of

  • Flexibility offered by NFV

Proposal

8 9/26/2014 Exploiting ICN for Flexible Management of SDN

We Propose Function Centric Service Chaining (FCSC) Key idea of information centricity => Name based forwarding Caching is a service that it facilitates

slide-9
SLIDE 9

FCSC Basic Design

9/26/2014 Exploiting ICN for Flexible Management of SDN 9 Forwarding Layer Ingress Middlebox Policy Module Routing Module Flow Features Next Hop Flow Layer Controller

(a) SDN

Router/Switch What Where

slide-10
SLIDE 10

FCSC Basic Design

9/26/2014 Exploiting ICN for Flexible Management of SDN 10 Forwarding Layer Ingress Middlebox Policy Module Routing Module Flow Features Next Hop Flow Layer

Naming Layer

Ingress Middlebox Policy Module Flow Features Name Routing Module Next Hop Controller

(a) SDN (b) FCSC

Router/Switch

These layers are already available in NDN/CCN enabled nodes

What Where What Where

slide-11
SLIDE 11

FCSC: Forwarding Engine

Per flow solutions Entries ∝ flows FCSC Entries ∝ functions

11 9/26/2014

Flow identifier FACE Flow-id 5 Tuples Application Function identifier FACE Function A Function B

Scalability Exploiting ICN for Flexible Management of SDN

slide-12
SLIDE 12

FCSC: Forwarding Engine

Per flow solutions Entries ∝ flows FCSC Entries ∝ functions

12 9/26/2014

Flow identifier FACE Flow-id 5 Tuples Application Function identifier FACE Function A X Function B Function A Y

Intrinsically supports the presence of multiple instances for the same functionality and can perform network layer load balancing among these nodes

Scalability Scalability Flexibility Reliability Exploiting ICN for Flexible Management of SDN

slide-13
SLIDE 13
  • Naming Strategy

– chain: DPI/cache/egress-R5

  • Scheme identifiers could be chain, monitor, ctrl

– Prefix popping (simple and stateless task) will be done at the

node hosting that particular network function instance

  • Routing Strategy

– Middleboxes advertise the prefix they are serving

  • prefix (similar to advertising the data they serve)

– Forwarding table is accordingly populated

  • Can use centralized/distributed routing schemes
  • Stateful Middleboxes

– In some cases, it is necessary to maintain states – Use chain:firewall/_A/cache instead of firewall/cache

FCSC Design Strategy

13 9/26/2014 Exploiting ICN for Flexible Management of SDN

slide-14
SLIDE 14
  • Flow Initiation

– Ingress (or via the controller) knows the list of functions a flow needs – The controller does not have to inform all the routers of the

forwarding rules for the new flow, just ingress(es) is sufficient

  • Proactive Rules

– Can set at ingress the set of functions a flow might require

  • Not required to proactively set forwarding states in routers
  • No need to set paths from each ingress for all flows [O(𝑂2)]

FCSC Architecture Description

14 9/26/2014 Exploiting ICN for Flexible Management of SDN Scalability

Flow identifier Functions Needed Flow-1 DPI, Cache Flow-2 DSA, TCP Opt. Flow-3

slide-15
SLIDE 15
  • Policy change by middleboxes

– Just change the name, i.e function list – No need to change forwarding rules in the routers

FCSC Architecture Description

15 9/26/2014 Exploiting ICN for Flexible Management of SDN Flexibility Scalability

slide-16
SLIDE 16

Lifetime of a Packet

9/26/2014 Exploiting ICN for Flexible Management of SDN 16 Ingress Egress Firewall A Load Balancer Firewall B R5 chain:/Cache/R5 chain:/LB/_FW/Cache/R5 Cache chain:/Firewall/_B/Cache/R5 chain:/Cache/R5 chain:/R5 DPI chain:/DPI/Cache/R5

DPI

Middlebox Prefix Popper ICN-Switch

slide-17
SLIDE 17
  • We use a custom simulator in Java

– Used in previous works such as COPSS[1], G- COPSS[2], Coexist[3]

  • Compare to a centralized controller based SDN

– Decentralized approaches exist

  • Inconsistent state can affect performance
  • Communication overhead to keep them in sync is not known

– Even if a decentralized approach exist, each controller will be

responsible for a set of routers (a portion of the network)

  • Topology (Demonstrate the benefits of FCSC)

– Synthetic topology – Real world topology

Evaluations

9/26/2014 Exploiting ICN for Flexible Management of SDN 17

slide-18
SLIDE 18

Synthetic Topology

  • Link latency

– between switches is 2ms – between switches and the end-systems (middlebox, src, dst, control)

is 10ms.

  • Bandwidth

– 100Mbps

  • The processing latency on all the middleboxes (including Ctrl)

is 1ms, or 1000pps (packets per second).

  • The sending rate at src is also 1000pps.

18 9/26/2014

R5 R4 R6 Ctrl DPI N1 [A] N2 [B] Dst Src N3 [A] N4 [B] R1 R2 R3

Exploiting ICN for Flexible Management of SDN

Focus: To understand how FCSC can compliment SDN

slide-19
SLIDE 19

Dynamic Failure Recovery

9/26/2014 Exploiting ICN for Flexible Management of SDN 19

FCSC looses less packets while recovering from failure

60 70 80 90 100 50 100 150 200 250 300

Latency (ms) Packet ID

FCSC SDN

slide-20
SLIDE 20

Dynamic Adaption to New Instances

9/26/2014 Exploiting ICN for Flexible Management of SDN 20

FCSC flows able to use new instances since routers are able to make forwarding decisions on a per packet basis

60 70 80 90 100 50 100 150 200 250 300

Latency (ms) Packet ID

FCSC SDN

slide-21
SLIDE 21

Dynamic Policy Change on Middleboxes

9/26/2014 Exploiting ICN for Flexible Management of SDN 21

FCSC routes via the new function at 30th packet itself

50 75 100 125 150 20 40 60 80 100

Latency (ms) Packet ID

FCSC SDN

slide-22
SLIDE 22

Large Scale Evaluation

  • Rocketfuel Topology

– Exodus AS-3967

  • 18 cities used as core network
  • Latency

– 30 links – Latency ranging from 2ms – 21ms (avg 6.6ms) – Latency between end-hosts, controller and core routers set to 6ms

  • Bandwidth is 100 Mbps
  • We assume that 11 different functions are required

– One is a DPI function that can rewrite the required function list

  • Each flow belongs to one of 100 different applications

– Each application requires a range of functions (1 to 4)

22 9/26/2014 Exploiting ICN for Flexible Management of SDN

Focus: Verify if benefits of FCSC hold true in the presence

  • f varying number of heterogenous flows and instances
slide-23
SLIDE 23
  • 100 long lasting flows (5 min)

– All flows start at 0s – Different sending rate (120Kbps to 1.05Mbps)

  • The simulation starts with one instance

– A new instance is launched at a random location every 30s

Varying Number of Function Instances

9/26/2014 Exploiting ICN for Flexible Management of SDN 23

75 85 95 105 1 2 3 4 5 6 7 8 9 10

  • Avg. Latency (ms)

# of instances per function FCSC SDN

slide-24
SLIDE 24
  • # of flows: 50 – 500
  • Arrival time: Random such that they start in the first 5 min

– Duration: 0.05s to 91.24s – Sending rate: 1.2 Mbps to 11.09 Mbps.

  • # of middlebox creation/failure events: 1,151
  • 95% confidence Interval

Varying Number of Flows - I

9/26/2014 Exploiting ICN for Flexible Management of SDN 24

50 60 70 80 90 100 110 50 100 150 200 250 300 350 400 450 500

  • AVG. LATENCY (MS)

# OF FLOWS FCSC SDN

slide-25
SLIDE 25

Varying Number of Flows - II

9/26/2014 Exploiting ICN for Flexible Management of SDN 25

0% 10% 20% 30% 50 150250350450 Packet loss rate # of flows FCSC SDN 4 8 12 16 50 150 250 350 450 # of rules (k) # of flows FCSC SDN

  • # of flows: 50 – 500
  • Arrival time: Random such that they start in the first 5 min

– Duration: 0.05s to 91.24s – Sending rate: 1.2 Mbps to 11.09 Mbps.

  • # of middlebox creation/failure events: 1,151
  • 95% confidence Interval
slide-26
SLIDE 26
  • First attempt of enhancing SDN with information centricity by

dealing with a common and important problem of network mgmt.

– Service Chaining

  • Simple solution, but the potential could be huge

Conclusion

9/26/2014 Exploiting ICN for Flexible Management of SDN 26

slide-27
SLIDE 27

As usual, open_issues/work_to_be_done outweighs work done 

Future Work

9/26/2014 Exploiting ICN for Flexible Management of SDN 27

Work on the more detailed aspects to improve naming, routing, forwarding and etc. But, also work to convince the non-ICN community: other SDN use-cases, prototyping, deployment in small scale, better performance evaluations.

slide-28
SLIDE 28

Contact: Arumaithurai@cs.uni-goettingen.de Acknowledgements Parts of this research was funded by the joint EU FP7/NICT GreenICN project, under EU grant agreement 608518 and NICT contract 167 and the Volkswagen Foundation Project “Simulation Science Center”.

Thank you for your attention

slide-29
SLIDE 29

1. COPSS: An Efficient Content Oriented Publish/Subscribe System, Jiachen Chen, Mayutan Arumaithurai, Lei Jiao, Xiaoming Fu, K. K. Ramakrishnan, ACM/IEEE Symposium on Architectures for Networking and Communications Systems (ANCS 2011), Brooklyn, NY, USA, October 2011. 2. G-COPSS: A Content Centric Communication Infrastructure for Gaming, Jiachen Chen, Mayutan Arumaithurai, Xiaoming Fu, and K.

  • K. Ramakrishnan, The 32nd IEEE International Conference on

Distributed Computing Systems (ICDCS 2012), Macau, China, June 2012. 3. Coexist: Integrating Content Oriented Publish/Subscribe Systems with IP, Jiachen Chen, Mayutan Arumaithurai, Xiaoming Fu, K. K. Ramakrishnan, ACM/IEEE ANCS 2012, ACM, October 2012.

References

9/26/2014 Exploiting ICN for Flexible Management of SDN 29

slide-30
SLIDE 30

Extra Slides

slide-31
SLIDE 31
  • Indirection based service chaining

– They rely on predetermined nodes that provide the service – Con: Poor routes; Cannot make good use of the flexibility of NFVs

  • Policy based routing

– Based on traffic characteristics (5-Tuple-IP; rate) – Con: Policy decision needs to be communicated early to the routers,

i.e., flow based forwarding rules (issues of scalability, flexibility)

Related Work - I

31 9/26/2014 Exploiting ICN for Flexible Management of SDN

slide-32
SLIDE 32

Lifetime of a packet

9/26/2014 Exploiting ICN for Flexible Management of SDN 32

Middlebox Prefix Popper ICN-Switch Ingress Egress Firewall A Load Balancer Firewall B R5 chain:/Cache/R5 chain:/LB/_FW/Cache/R5 Cache chain:/Firewall/_B/Cache/R5 chain:/Cache/R5 chain:/R5 DPI chain:/DPI/Cache/R5 DPI

slide-33
SLIDE 33
  • What if we want all the packets of a flow to go to the same

instance?

– Use name associated with that particular instance

  • E.g, chain:firewall/_A/cache
  • What if we want the packets in both directions to go through

the same instance?

– Again, use name associated with that particular instance

FCSC special cases

9/26/2014 Exploiting ICN for Flexible Management of SDN 33

slide-34
SLIDE 34
  • What is the Impact of the choice of a routing scheme?
  • Decision affects the flexibility and reliability

– Centralized controller based scheme

» Global knowledge

– Distributed routing scheme

» Need to synchronize better » But information can propagate to neighbours quickly

  • Possibility to use a load-balancer middlebox instance if required

FCSC Routing

9/26/2014 Exploiting ICN for Flexible Management of SDN 34

slide-35
SLIDE 35
  • FCSC therefore

– Flexibility

  • Enables switches to dynamically detect load and accordingly

instantiate/dispose Network functions

  • Allows flows to use newly instantiated NFVs and faster recovery from

node/link failures

  • Allows dynamic modification of the functions needed by a flow on the

controller or middleboxes

– Scalability

  • By placing the flow state in the packet header, the number of states stored in

the network is reduced

– Therefore more scalable than the per-flow state solutions

  • Intrinsically supports the presence of multiple instances for the same

functionality and can perform network layer load balancing among these nodes

– Reliability

  • Allows faster recovery from node/link failures by using the forwarding table to

find another instance

FCSC Advantages

9/26/2014 Exploiting ICN for Flexible Management of SDN 35

slide-36
SLIDE 36

Proactive rule for flow initiation

9/26/2014 Exploiting ICN for Flexible Management of SDN 36

60 70 80 90 100 10 20 30 40 Latency (ms) Packet ID FCSC SDN