End to end OBS Applications

End to end OBS Applications

Dan StevensonVP, Advanced Network and Information

Technology [email protected]

E-science Application Needs

• Data Grid– Multi-tera Byte data sets– Moved between Labs and supercomputers

• Compute Grid– Also requires low latency communications

• LambdaGrid– Moved over dedicated wavelengths– No routers in path– Sparse network– Rapid (sec-min) wavelength provisioning

CACM Nov. 2003, SPECIAL ISSUE: Blueprint for the future of high-performance networking

Introduction, Maxine D. Brown, pp 30 - 33

TransLight: a global-scale LambdaGrid for e-science, Tom DeFanti, Cees de Laat, Joe Mambretti, Kees Neggers, Bill St. Arnaud pp: 34 - 41

Transport protocols for high performance, Aaron Falk, Ted Faber, Joseph Bannister, Andrew Chien, Robert, Grossman, Jason Leigh pp: 42 - 49

Role of OBS

• Ultra-fast light path provisioning via OBS– Milliseconds with MEMS micro mirrors– Microseconds with electronically activated switches

• JIT OBS as a grid service– Establish connections through an GSI-authenticated,

OGSI-based Grid Service interface

• OBS interfaces for Grid tools• Transport protocol optimized for JIT OBS

Grid Interface

gridappication

OGSA

GRID-JIT

JIT OBS

data

gridappication

OGSA

GRID-JIT

JIT OBS

data

JITPAC

OXC

sig sig

• Permits scheduling of channels as Grid resource

gridFTP

• Linux client to Linux client

• With JIT OBS network adapter

gridFTP

ST

JIT OBS

data

gridFTP

ST

JIT OBS

data

JITPAC

OXC

sig sig

JIT OBS Network Adapter

JITPAC Evolution

• TL1 Interface– Generic OXC and PXC control interface

• Soft JITPAC– All software JIT implementation on PC– Motivated by cost considerations– Deployment on MEMS photonic switches

• JITPAC II– Higher performance

•160,000 messages per second– On board electronically activated switches

•SOAs, PLZT, EO Polymer•10 micro-second bursts

– Limited wavelength conversion– 4 fiber pairs, 4 wavelengths each– Targeting deployment to NLR

Proposed Equipment Stack

• Avoids difficult physical layer issues

• Permits experimentation

• With burstification and end to end OBS

JITPAC V2with SOA

switch

CISCOIP switch

CISCOOXC

Linux Cluster

OBS NA

JITPAC V2with SOA

switchIP switch

COTSOXC

Linux Cluster

OBS NA

Bur

stifi

er

Physical Layer Challenges

• Signal– Burst mode synchronization– Signal quality monitoring in network– Variable power levels at receiver– Rate adaptation

• Our approach– Long haul: use OEO and commodity systems– LAN and Metro: reduced problem set, need conversion– Separate research focused on physical layer

OXC(Cisco 15454)

1550 nm 1550 nm

850 or 1310 nmGbE

JIT Security and Authentication Extensions

• User must “pay” for the use of network resource from a limited pool of credits.

• Solve two problems at once:– Accountability for use of network resources– Verified identity of the user

• Ability to perpetrate DOS attacks is greatly diminished by limiting attacker’s resources

• Similar to DARPA supported ARQoS project– Used economic models from telephony, applied them to

DiffServ (resource reservation framework for IP networks)

Routing Protocol

• Data path (with QoS) & signaling path routing• Intra & inter – domain issues• Routing loop prevention• Avoidance of switch to switch communications

Medusa Emulator

• Three major pieces:– Automatic Network Topology Generator– GUI Based Network Editor– Medusa Signaling Testing Part

• Major Features– real-time traffic forwarding for signaling testing– script tools to randomly generate large scale network config file – import from and export to xml network config file– Editing network nodes, ports, links– export the xml config file for each node as to given xml schema– pull down specific interface/port or simulate certain drop rate– Platforms

• Java-based, Windows or UNIX• References

– J-Sim (formerly known as JavaSim): a component-based, compositional simulation environment http://www.j-sim.org/

– BRITE: an approach towards universal topology generation http://www.cs.bu.edu/brite/

Medusa Screen Shot

Backup chartsBackup charts

Problems Facing OBS

• Technical– Immature conversion technology – Physical layer issues

• Electro-political– Momentum of GMPLS “inventing another protocol is

irresponsible”– Violated expectations

•Optical bubble•Over hyping promise of optics

– Turner effect•“Optical networking is dead.”

Wavelength Converter

Network Adaptor

ATDnet Testbed

• DC metro area• 8 channels• ATM overlay• OBS overlay

– DIA, NRL, LTS

DIA

DISA

DARPA

NASA

LTS

NRL

all fibers carry8 wavelengths

data rates vary between2.5 Gbps and 40 Gbps

depending onexperimental activity

https://www.atd.net/Boling AFB Washington, DC

Adelphi, MD

Baby picture goes here

JITPAC Test Lab