Impact of Path Performance on Concurrent Multipath Transmission - - PowerPoint PPT Presentation

Impact of Path Performance on Concurrent Multipath Transmission within Routing Overlays

• K. Tutschku, T. Zinner, A. Nakao

8thWürzburg Workshop on IP: Joint EuroNF, ITC, and ITG Workshop on "Visions of Future Generation Networks“ (EuroView2008)

Network Architecture Group – Network Virtualization Lab

Future Internet? How to approach?

Clean slate approach?

UC Berkey, Intel: Declarative Networking; Parc

(V. Jacobson) : Content Centric Networking

Evolutionary approach?

IRTF RRG/ ALTO BOF: Location / Identifier Split,

(ISP-Supported) Path Selection Service

iPhone 3G start: 3G already commodity but Apple’s App Store dominates blogs  FI should please applications

What Features are Currently Missing?

End-to-end QoS, but:

QoS-Islands are available (depending on

technology and provider)

Application layer mechanisms are highly

efficient (Skype, P2P CDNs)

Reliability, but:

Sophisticated TE and Resilience concepts

available at experienced ISPs

Competition/Business models, but:

• J. Crowcroft: “… i can go on the web and get my

gas, electricity, … changed , why is it not possible to get a SPOT price for broadband internet?” (E2E-interest mailing list on April 26th, 2008)

 Why is it so complex to include such features in the Internet?

Capabilities and Features of the Current Internet:

(Special Thanks to Michael Menth for visualization support)

PoP-level Path Diversity

• R. Teixera et al. (2003): 50% of PoPs

have more than 3 links to other PoPs  Multiple routes readily available!

Use neighbors that minimize delay!

Easy in an Euclidean space PlanetLab measurements (S.

Banerjee et al., 2004): 25% of TI violations  Better routes exist

A C B

Triangle Inequality (TI)

D(A,C) ≤ D(A,B) + D(B,C)

 Apply routing overlays for integration and push application layer mechanisms to routing layer

SORA: Scalable Overlay Routing Architecture

One-hop source routing

Gummadi et al (2004):

Increased reliability while being scalable

Nakao, Lane (2007):

Multi-path aware  application to resource mgmt C-D separation implemented (path computation and forwarding) Pushing End-to-End Principle to Routing

ᴉ

May be inefficient  Reduction of overhead (since edge-based)

1 Divert selected endhost packets 2 Request Paths for Diverted Packets 3 Encapsulated, send using path 4 Decapsulate, egress to destination

Network Virtualization: Capabilities

Computational Virtualization

Use Computer Center as a PC with

user-specific environment

Fully Virtualized Networks

Use network with user-specific

mechanisms (routing/resource access)

Network-virtualization capable

router everywhere

Resource isolation (I/O virtualization)

and resource independence (resource can be located every where)

Virtual topology and virtual resource

management

Hardware HostOS Apps VM VM GuestOS GuestOS Apps Apps

Hardware VMM VM VM GuestOS GuestOS Apps apps VM MngOS Manager

Hypervisor-based Host-based

Decision (e.g. by application) Data Discovery Dissemination Network-view Direct control Adapted from A. Greenberg,

• J. Rexford et al. (2005)

EuroView architecture (a first concept, UniWürzburg)

Application Mediation Connectivity

Virtual Resource Management: Concurrent Multipath Transfer in Multihomed Overlays (Striping)

Physical topologies

• f different

providers Different overlays

(First) aim: Obtain high throughput Solution: Combine multiple paths (provided by disjoint overlays) into an overall transport pipe

• verall transport

pipe

Individual Overlay paths

Features:

Increased reliability and very high

capacity

Interdomain traffic management and

edge-based performance control

avoid congested areas conceal triangle inequality violations exploit high access bandwidth discover available bandwidth in

core

Transmission Model

p1,1

dst

Assump7on: use k parallel paths on m

• verlays

p1,n1 pm,1 pm,nm

src

k paths With paths Data stream divided at router into segments with k parts 1 k 2 k parts have arrived k parts are send in parallel at 7me t k‐1 each provider will offer a set niof parallel paths (i = 1…m)  

k 1

• verlay 1
• verlay m

 Re-ordering?

Performance Evaluation

First analysis by Y. Nebat and M Sidi (2006)

Two synchronous, equal capacity paths (= sources) Maximal packet delay dmax= 300msec; Uniformly delay Normal distributed delay: µ = 150msec; σlow = 50 msec and ; σhigh = 75 msec

Delay Buffer P(d) P(K) Packet delay d [msec] Resequencing buffer occupancy K

norm: σlow / σlow norm: σlow norm: σhigh uniform uniform / uniform norm: σhigh / σhigh uniform / norm: σlow

Conclusion

Relationship to P2P content distribution:

View each path as a peer Select paths/peers such that reordering/throughput/resilience is

minimized/optimized Network Virtualization

Enables multiple, parallel application-specific routing overlays

(specific i.e. wrt. routing/topology/resource management ) Future work:

Identify timescale for overlay setup / adaptivity Extend analytical performance model to N paths with hetero-

geneous capacity (Paper available in September 2008 from NICT, University of Wuerzburg, and University of Vienna)