OptIPuter Middleware Software Implementation Plans

1
OptIPuter Middleware Software Implementation Plans

• Nut Taesombut
• Department of Computer Science and Engineering
• University of California San Diego
• optiputer_at_csag.ucsd.edu
• 4th OptiPuter Backplane Architecture Workshop
• January 17, 2006

2
Objectives

• Tame Configurable Optical Network for
  Applications
• Define application abstractions Model of use
• Define software architecture Internal, external,
  infrastructure
• Deliver the Communication Capabilities of
  Lambdas
• High-performance transports
• Novel communication capabilities Multi-endpoint,
  photonic multicast
• Expose and exploit optical network control
• Enable and Demonstrate Visualization and
  Data-Intensive Applications
• Interactive communication, novel
• Direct access to distributed display and storage
• Aggregate wide-area distributed storage
  efficiently

3
OptIPuter System Software Architecture
Distributed Applications/ Web Services
Visualization
Telescience
SAGE
JuxtaView
Data Services
Vol-a-Tile
LambdaRAM
PIN/PDC
4
Distributed Virtual Computer (DVC) Grid
Resources Private Network

• Tame Configurable Optical Network for
  Applications
• Leverage Grid resource acquisition and use model
• Integrate and optimize network configuration and
  resource selection
• Provide a Simple Performance Abstraction and
  Environment for Applications
• Access resource in SAN-like fashion (private
  connection, deterministic performance
• Utilize standard and custom protocols (e.g., GTP,
  UDT, RBUDP) in uniform way

5
OptIPuter High-Performance Transport Protocols

• Bridge the Gap between High Speed Link
  Technologies and Growing Demands of Advanced
  Applications
• TCP has well-documented performance problems on
  long-haul networks
• Pursue complementary avenues of investigation
• Efficient congestion/flow management, fairness
  among flows
• High-speed group communication (multipoint-to-poin
  t, multipoint-to-multipoint)

6
Unified Communication Framework
Distributed Applications/Web Services
DVC Communication Service/API
Socket-like API
File Transfer API
DVC Conf. API
Name resolution, Group Comm. Management
DVC Core Service
DVC Descriptor (Comm. Config.)
?- stream
GTP
UDT
XCP
RBUDP
CEP

• Provide Unified Framework for Access to a Range
  of Transport Protocols/Mechanisms
• Single implementation required for the
  application
• Adaptable to diverse resource conditions
• Configurable transport protocol use
• Choose the best performer for your environment

7
OptIPuter System Software Year 2005

• OptIPuter System Software Integrated and
  Delivered as Rocks Rolls
• Bootable CD and Roll CDs for entire cluster
  building
• Rolls are optical extension to basic cluster
  configuration
• OptIPuter System SW Roll (Gold Roll) v.1
  July2005
• OptIPuter Software Architecture
• DVC Middleware v1.0
• Basic security, namespace and job management
• Coordinated resource and network selection
• Automated resource and network configuration (via
  PDC, Globus)
• Uniform access to novel communication protocols
  (GTP, CEP)
• Advanced Transport Protocols
• GTP v0.95 Receiver-based rate allocation and
  adaptation
• CEP v1.1 Multi-endpoint file transfer,
  greedy-based flow scheduling
• Optical Network Configuration
• PDC v2.0 Intra-domain lambda reservation and
  allocation

More information on Rocks http//www.rockscluster
s.org/
8
OptIPuter 5-layer Demonstrations iGrid2005

• Scientific Applications (Neuroscience,
  Geoscience)
• Visualization
• Distributed Virtual Computer (DVC)
• Novel Transport Protocol (GTP, lambda-stream)
• Optical Network Configuration (PIN/PDC)

9
OptIPuter Software Stack
10
Cross Team Integration and Demonstrations

• TeraBIT Juggling, 2-layer Demo SC2004, Nov2004
• DVC middleware, high-speed transport (GTP)
• Move data between OptIPuter network endpoints
• Endpoints across UCSD, UvA, UIC, Pittsburgh
• Achieved 17.8Gbps, a TeraBIT in less than one
  minute
• 3-layer Demo AHM2005, Jan2005
• Visualization (JuxtaView/LambdaRAM), DVC
  middleware, high-speed transport
• Remote data visualization (visualization NCMIR
  storage UIC and UvA)
• Use DVC to establish visualization environments
• Automated Grid resource selection and binding
• Achieved 2.6 Gbps on 7 Streams
• 5-layer Demos iGrid2005, Sep2005
• Applications, visualization, DVC middleware,
  high-speed transports (GTP), optical network
  configuration (PIN/PDC)
• Demo 1 Collaborative Data Visualization with
  Earth-Sciences
• Demo 2 Real-time Brain Data Acquisition,
  Assembly and Analysis

11
Collaborative Data Visualization with
Earth-Sciences
Amsterdam
Seattle OptEx
Chicago OptEx
10 GE
10 GE
Chicago
UCSD OptEX
10 GE
20 GE
San Diego

• Scientific Collaboration with Parallel
  Interactive 3D Visualization
• Geographically distributed storage (San Diego,
  Chicago, Amsterdam)
• Multi-gigabyte datasets (25GB)
• Visualization centers at Calit2 and SIO/UCSD
• Demonstrating OptIPuter Technologies
• Dynamic Grid resource and network configuration
• Real-time data acquisition with GTP, Achieved
  16.3Gbps peak rate (81.5 link utilization)
• From UvA, UIC and JSOE to Calit2

12
Using DVC

• User Specifies Resource and Network Connectivity
  Requirements
• DVC Service Selects a Matching Resource
  Configuration
• DVC Service Allocates Grid Resources and Private
  Networks

DVC Service
Grid Resource Manager PIN/PDC Service
Resource Spec.
UIC Cluster
Chicago OptEx
OXC
Seattle OptEx
UvA Cluster
OXC
SDSC Cluster
San Diego OptEx
OXC
CSE Cluster
NCMIR Cluster
Calit2 Cluster
JSOE Cluster
13
Multi-scale Correlated Microscopy Experiment
Reduction of information across scales and
imaging technologies
Simultaneous visualization of multi-scale biologic
al specimens and high-resolution video-conferencin
g
Source David Lee, NCMIR http//www.c5d.org/Projec
ts/iGrid/igrid05.html
14
Real-time Brain Data Acquisition, Visualization
and Collaborative Analysis
Japan
Chicago
HV Electron Microscope
Seattle OptEx
Chicago OptEx
10 GE
UIC
KDDI/Osaka
storage
10 GE
UCSD OptEX
Light Microscope
CalIT2/UCSD
NCMIR/ UCSD
10 GE
20 GE
viz. cluster
viz. cluster
San Diego

• Transparent Operation of a Scientific Experiment
• Visualization of multi-scale sample specimens
  live videos
• Globally distributed visualization, storage and
  network resources
• Real-time data acquisition and visualization
• Demonstrating OptIPuter Technologies
• Visualization SAGE, TeraVision, JuxtaView,
  Vol-a-tile, LambdaRAM
• Dynamic resource and network configuration DVC,
  PIN/PDC
• High-speed transports LambdaStream

Source David Lee
15
Snapshot of SAGE on 100-Megapixel Tiled Display
at iGrid2005
JuxtaView displays a dataset stored remotely at
UIC/Chicago. The dataset is interactively
streamed using LambdaRAM
A real-time video of a light microscope from
NCMIR streamed using TeraVision
An HDTV camera feed over TeraVision shows a
conference room at NCMIR
Pre-recorded video stream from Osaka, Japan shows
an electron microscope experiment
Dataset acquired using an electron microscope
Vol-a-Tile rendering a volume dataset stored
locally
Source Raj Singh
16
Plan for OptIPuter Year 4
17
OptIPuter System Software Year 2006

• DVC Middleware v1.2
• Improved resource selection algorithms
• High-speed communication with UDT and ?-lambda
• Web service interface (WS-RF)
• GTP v1.3
• Capability management at both senders and
  receivers
• Improved CPU efficiency and scalability
• CEP v2.0
• Improved performance and stability
• Integration of advanced protocols (e.g., GTP)
• UDT v3.0 (aka Composable-UDT)
• Support for multiple high-speed congestion
  control algorithms
• Congestion controlled unreliable messaging
• ?-Stream
• Support for multipoint to multipoint communication

18
OptIPuter Software Summit Rolls

• OptIPuter Software Summit Jan2006
• Integration of end-to-end OptIPuter software
  intensive testing
• OptIPuter Software Summit Rolls
• OptIPuter System Software Roll v.2
• DVC Middleware v1.0.1
• DVC 1.0 integration with PIN
• Advanced Transport Protocols
• GTP v0.95, CEP v1.1, UDT v2.0
• Optical Network Configuration
• PDC v2.0, PIN v0.3
• OptIPuter Visualization Roll (OptiViz) v.1
• Visualization Toolkit
• Ethereon, JuxtaView, Vol-a-Tile
• Data Toolkit
• LambdaRAM

19
Plan for GT4.0 Integration
20
Exposing OptIPuter Capability as Web Services

• OptIPuter Software System is an Service-Oriented
  Architecture
• Capabilities are presented as web services
  instantiated and accessed
• Distributed Virtual Computers (DVCs), Real-Time
• Network Protocol Configuration/Management
• Optical Lightpath Configuration
• Well-defined interfaces (i.e., WSRF)
• Services are Linked and Accessed by Clients
  through the DVC Session Manager

DVC Core
Real-Time DVC
WS-RF
DVC Session Manager
Comm. Config.
Optical Network Configuration
GT4/GRAM
Web Service Client
Grid Resource Manager
21
Example of Use

• User requests resources and connectivity via WS
  client

WS-RF
DVC Session Manager

• DVC service finds a matching resource
  configuration

DVC Core
DVC Grid Resource Binding
DVC Net Binding

• Grid resources and network are allocated using WS
  interfaces (OGSA, WS-RF) DVC is created

Configurable Optical Network

• User obtains DVC reference (WS endpoint
  reference) and submit computation/communication
  tasks through it

22
Question?