Office of Science MICS Division Department of Energy High-Performance Networking Research Program...

Office of ScienceMICS Division

Department of Energy

High-Performance Networking Research Program

Program Manger: Thomas D. NdousseTel: 301-903-9960

Email: [email protected]

Project Quad Charts

Bandwidth Estimation: Methodologies and Applicationsk claffy, CAIDA at SDSC & Constantinos Dovrolis, Univ. of Delaware

Brief Summary of the Project • Task 1: Develop accurate, fast, and non-intrusive bandwidth estimation (bwest) methodologies and measurement tools.• Task 2: Compare and evaluate different bwest tools (both for end-to-end and per-hop bandwidth metrics), characterizing any observed errors.• Task 3: Use bwest methodologies in transport protocols and applications to optimize throughput for high bandwidth-delay-product paths.• Task 4: Prototype bwest middleware to monitor performance between network domains in real-time.

04/21/23 04:16 PM

High-Performance Network Research SciDAC Project

MICS Program Manager: Thomas Ndousse

Innovative end-to-end probing techniques to measure capacity (max possible throughput in empty path) and available bandwidth (max throughput under current load):

– Packet Train Dispersion (PTD).– Variable Packet Size (VPS).– Self-Loading Periodic Streams (SLoPS)

Methodologies to check for overbuffered or underbuffered network paths. Smooth pacing in TCP, driven by bwest measurements. Smooth bwest driven rate-control for UDP-based applications.

The Novel Ideas

Compare and evaluate existing bwest tools: - Hop-by-hop tool survey Jun01 - Aug02 - End-to-end tool survey Jun 01 - Jun02 Bandwidth measurement middleware - Create/maintain testbed Jun01 - Jun04 - Collect link characteristics Jun01 - Jun04 - Correlate active/passive measurements Jun01 - Jun04 Capacity estimation tool (pathrate) v2.1.2 Dec01 DONE - Add GUI to aid analysis of results Dec02 Available bandwidth tool (pathload) Mar02 - Paper at PAM’02 Mar02 Develop UDP-based rate-controlled file transfer app driven from bwest measurements Dec02 Real-time path monitor using bwest middleware Dec03

Milestones/Dates/Status IMPACT: Allow scientific applications (transferring terabytes of data) to efficiently use high-performance networks .

– Use explicit bwest measurements instead of implicit bwest via TCP’s congestion control algorithms.– Provide easy-to-use tools for monitoring network path performance.

CONNECTIONS: – Apply bwest methodologies to Web100 and Net100 projects.– Correlate bwest to loss/delay (e.g. PingER project)– Establish prototype bwest middleware in ESnet and for DOE labs and investigators.

Impact and Connections

Security and Policy for Group CollaborationSteven Tuecke, Argonne National Laboratory, Carl Kesselman, USC Information Sciences Institute

Miron Livny, U. Wisconsin, Madison

Impact and Connections IMPACT: We expect this project to result in:

Standardization of new PKI-based approaches to credential management, restricted delegation, policy management

Development of security tools and services for collaboration Widespread deployment and adoption of approaches and tools

CONNECTIONS: This work builds on the Globus Toolkit’s widely used Grid

Security Infrastructure (GSI), and will be in future Globus Toolkit. To be used by numerous SciDAC collaboratories, including DOE

Science Grid, Particle Physics Data Grid, Earth Systems Grid, and Fusion Collaboratory

Also to be used by many non-DOE projects worldwide, including NSF PACI DTF, NASA IPG, and European Data Grid

Milestones/Dates/Status Demonstrate CAS prototype @ SC’01 November 2001 Complete X.509 & GSS standards drafts February 2002 Deliver draft standard conforming GSS April 2002 Deliver CAS w/ simple policies May 2002 Demonstrate Online CA & CR September 2002 Complete Online CA & CR standards drafts December 2002 Finalize X.509 & GSS standards February 2003 Deliver Online CA & CR March 2003 Deliver CAS w/ rich policy & app support May 2003 Finalize Online CA & CR standards December 2003 Deliver standards-based Online CA & CR March 2004 Deliver CAS w/ accounting support May 2004

The Novel Ideas Enable collaborative work, with common security tools that address: - Large, geographically & organizationally distributed membership - Membership with diverse expertise, comprising different roles - Community resources with associated community policies Develop novel tools and approaches for: - Management of collaboration membership and resources - Online CA & Credential Repository (CR), local security integration - Management of roles and privileges - Community Authorization Service (CAS), restricted delegation - Integration into collaborative tools and environments

September 2001MICS Program Manager: Thomas Ndousse

CAS1. CAS request, with resource names and operations

Does the collective policy authorize this

request for this user?

user/group membership

resource/collective membership

collective policy information

Resource

Is this request authorized for

the CAS?

Is this request authorized by

the capability? local policy

information 4. Resource reply

User 3. Resource request, authenticated with

capability

2. CAS reply, with and resource CA info

capability

Community Authorization Service


INCITE: Edge-based Traffic Processing and Inference for High-Performance Networks Richard Baraniuk, Rice University; Les Cottrell, SLAC; Wu-chun Feng, LANL

Impact and Connections

IMPACT: Optimize performance of demanding applications

such as remote visualization and high-capacity data transfers

New understanding of the complex dynamics of large-scale, high-speed networks

New edge-based tools to characterize and map network performance as a function of space, time, application, protocol, and service

CONNECTIONS: Rice/SLAC/LANL synergy, SciDAC

Milestones

Analysis, modeling, and inference Multifractal, wavelet, tomography theory ongoing Traffic analysis toolbox 12/02 Passive path inference and tomography algs 10/03

PingER Add tomography, chirping, fat boy 04/02 Port extended PingER to Rice/LANL 10/02 Add new inference algs to PingER-NG 06/03 Evaluate, port PingER-NG to GIMI/NMF 04/04

MAGNeT / TICKET MAGNeT, TICKET (alpha distribution) 10/02 High-speed, high-utilization traffic traces 09/02 MAGNeT (public availability) 06/03

INCITE Summary

• Task 1: Multiscale traffic analysis and modeling

• Task 2: Inference algorithms for network paths and links

• Task 3: Network tomography

• Task 4: Active network measurement: PingER

• Task 5: Passive network Measurement: MAGNeT, TICKET

• Task 6: Passive path monitoring and tomography toolkit

Date Prepared: 10 Jan 02



incite.rice.edu

INCITE Novel Ideas

• Multiscale / multifractal analysis for traffic bursts

• Efficient “packet chirp” and “fat boy” path probing

• Active and passive network tomography

• Monitor for Application-Generated Network Traffic (MAGNeT)

• Traffic Information Collecting Kernel with Exact Timing

(TICKET)

• Augmented PingER

Logistical NetworkingPIs: Micah Beck, Jack Dongarra, James S. Plank / Tennessee; Rich Wolski / UCSB

Impact and Connections IMPACT:

Improved performance and scalability of data-intensive distributed application

Greater ease of and lower cost of deployment of new wide area data management strategies

Dramatically improved flexibility in data-intensive collaboration

CONNECTIONS: SciDAC: Net100, Data Grid, Scalable Systems, Data Mgt,

Computational Science (e.g. Climate, Supernovas) Base:Network Monitoring, Data Grid, Transport Protocols,

Storage Res. Mgt., IQ-Echo,

Milestones/Dates/Status–IBP applications demonstrated at SC’01–exNode support in NetSolve –Reliability/performance coscheduling alpha–Allocation policy simulation – Initial generalized caching infrastructure–Initial logistical overlay network on ESNet –Wide-area logistical peering mechanisms and policies–Resolution for highly volatile storage resources –Experimental IBP architectures–Large scale measurement and simulations

Novel Ideas Storage is too cheap to hoard. Storage can be a scalably shared network resource. Logistical Networking gives applications and middleware uniform control over buffering and routing of data. Data storage and data transport can be viewed as points on a spectrum of data management mechanisms. Monitoring and prediction can replace reservation as a means of scheduling storage resources. End-to-end networking principles can apply to storage.

Logistical Networking: Developing a communicative infrastructure with persistence

Tasks: -develop/deploy network storage depots -develop layered storage stack & tools -develop/validate scheduling techniques -optimize application performance

loci.cs.utk.edu

Date Prepared: 1/10/02



6-12mos

12mos

12-18mos

18-36mos

Net100PIs: Wendy Huntoon/PSC, Tom Dunigan/ORNL, Brian Tierney/LBNL


increase throughput of bulk transfers over high delay, bandwidth networks (like DOE’s ESnet)

select optimal paths and transport parameters for distributed (Grid) application (e.g.: GridFTP)

provide network performance data base from active and passive monitoring

CONNECTIONS: SciDAC: Astrophysics, Bandwidth Estimation, Data Grid,

INCITE, Logistical Networking Base:Network Monitoring, Data Grid, Transport Protocols

Milestones/Dates/Status Network probes and sensors Mon/Yr DONE - initial sensor and tool deployment 12/01 12/01 - data base design 4/02 - initial data base implementation 9/02 - final sensor/data base 6/03

• Transport protocol optimizations - protocol analysis 11/02 - initial tuning daemon 3/02 - bulk transfer tuning demos 8/02 - final tuning daemon 6/03 Multipath support - analytical analysis 8/02 - proof-of-principal routing daemons 12/02 - grid applications demos 4/03

Net100 Novel Ideas Net100 will tune network-UNaware applications based on recent and current link characteristics Net100 will tune more than just transport buffer sizes, such as

TCP AIMD parameters DUP threshold Delayed ACK

Net100 will determine optimal paths and whether to use multiple streams and/or multiple paths Net100 kernel utilizes passive monitoring from the Web100 kernel

NET100: Developing network-aware operating systems

Tasks:

-develop/deploy network probes/sensors -develop network metrics data base -develop transport protocol optimizations -develop network-tuning daemon

www.net100.org


High-Performance Network Research - Base Project


Self-Configuring Network Monitor (SCNM)PIs: Brian Tierney/LBNL and Deb Agarwal/LBNL


Build a monitoring infrastructure that will aid in debugging of distributed application communication and support both active and passive monitoring

CONNECTIONS: SciDAC: Net 100, DOE Science Grid, Astrophysics,

Bandwidth Estimation, Data Grid, INCITE, Net100 Base:Network Monitoring, Data Grid, Transport Protocols

URL: www-itg.lbl.gov/Net-Mon/Self-Config.html

Milestones/Dates/Status Monitor Daemon Year - Design base passive monitor daemon 1 - Activation mechanism integration 1 - Improvements to network drivers 1 - Improvements and enhancements to sensor mechanism 2 & 3

• Activation Mechanisms - Design basic activation mechanism 1 - Develop and deploy full activation capabilities 2 & 3 Results Handling Infrastructure - TCP dump viewing capabilities 1 - Develop improved data viewing capabilities 2 & 3 Deployment of Monitors - Deployment to initial ESnet sites (gig-E) 1 – 3 - Work with applications 2 & 3 - Additional ESnet sites 2 & 3

Novel Ideas• A secure monitoring infrastructure that applications can use to monitor performance of their own data streamsPassive – introduce traffic only in the form of monitoring data and requests for monitoring

Tasks Involved Develop a monitor activation mechanismDevelop monitor software and hardwareDevelop data collection and display capabilitiesDeploy monitorsWork with applications

SCNM: Developing a distributed passive network monitoring system


High-Performance Network Research Base Project


High-Performance Transport ProtocolsPI: Wu-chun (Wu) Feng, Los Alamos National Laboratory and The Ohio State University

Impact and Connections IMPACT.

Dynamic Right-Sizing- Auto-tuned, order-of-magnitude increase in throughput.- Vendor adoption, e.g., IRIX, Linux (still in the works)- Potential integration into GridFTP, Web100, Net100.

RAPID- Sliding reliability semantics may result in adoption of

RAPID by LANL large-data visualization team. CONNECTIONS.

Dynamic Right-Sizing: Web100, Net100, DOE Science Grid, Particle Physics Data Grid, Earth System Grid II,

RAPID: The LANL large-data visualization team, previously sponsored by the DOE NGI Corridor One project. Others?

Milestones/Dates/Status

Mon Yr DONE Simulation: Flow-Control Adaptation with Dynamic Right-Sizing

-Protocol Analysis & Design (ns-2) 12/01 12/01-Protocol Testing & Evaluation (rudimentary) 03/02 beta testing

Implementation: Flow-Control Adaptation with Dynamic Right-Sizing -Kernel Space, Linux 2.4.x 07/02 beta testing -User Space, drsFTP 01/03 alpha testing

-Protocol Testing & Evaluation (rudimentary) 03/03-Potential Integration with GridFTP 04/03-Deployment (kernel- & user space) 07/03

Simulation: RAPID-Effect of packet spacing 03/02preliminaries-Definition of API to middleware 03/02preliminaries-Sliding reliablity 07/03

The Novel Ideas Dynamic Right-Sizing: TCP Flow-Control Adaptation

for Grids & the Next-Generation Internet Automatically enhance network performance over the

WAN by as much as an order of magnitude while abiding by TCP semantics.

RAPID: Rate-Adjusting Protocol for Internet Delivery Provide smoother QoS support over the best-effort

Internet for grids and NGI while minimizing the need for widespread deployment of DiffServ or IntServ.

Goal: To significantly improve network performance in support of all computing environments, particularly grids and NGI. • TCP/IP Make the network fast but TCP friendly.

• Eliminate TCP’s flow-control bottleneck by automatically tuning buffer sizes.

• RAPID Make the network more adaptable.• Smooth QoS support over a best-effort

network.• User-settable reliability, providing a spectrum

of QoS from unreliable UDP to reliable TCP.

January 16, 2002

High-Performance Network Research - Base Project

MICS Program Manager: Thomas Ndousse-Fetter

IQ-EChoPIs: Schwan, Ahamad, Eisenhauer, Yalamanchili -- Georgia Institute of Technology

Impact and Connections IQ-ECho IMPACT.

– enable network-aware adaptable applications – cross-layer information exchanges will make effective

runtime tradeoffs in quality vs. performance across the protocol, middleware, and application levels

– enable the creation of efficient and adaptable Grid data services

CONNECTIONS:– Remote visualization (Supernova Visualization), source-

based filtering (Oakridge), program monitoring and steering– Extensible cluster platforms (NSF, DOE)– Remote sensing, monitoring, and security (DARPA, NSF)

Milestones/Dates/Status Year 1 Mon Yr DONE • performance attributes in ECho middleware 4/02• select and implement sample application 6/02• create instrumentation for performance attributes 8/02 Year 2 • evaluate and tune middleware 3/03• enable application for adaptation 3/03• extend/create configurable network protocols 6/03Year 3• integrate ECho-IQ with access grid software 3/04• demonstrate benefits in access grid environment 6/04

IQ-ECho – Interactive Quality of Service Across Heterogeneous Hardware/Software




• integrated QoS management through quality attributes • dynamic code generation relocates application-level

functionality to the most appropriate location• configurable protocols and kernel-level monitoring provide the

system-level support required for online quality management • vertical programming allows extending platforms while

programming applications

IQ-ECho Novel Ideas

represent information flows as event streams in event-based IQ-ECho middleware

use dynamic code generation to migrate application-level filtering/ data processing to appropriate network locations

use network-level feedback to drive application-level quality of service adaptations.

http://www.cc.gatech.edu/systems/projects/IQECho

PingERPIs: Les Cottrell SLAC


increase network and Grid application bulk throughput over high delay, bandwidth networks (like DOE’s ESnet)

provide trouble shooting information for networkers and users by identifying the onset and magnitude of performance changes, and whether they appear in the application or the network

provide network performance data base, analysis and navigateable reports from active monitoring

CONNECTIONS: SciDAC: High Energy Nuclear Physics, Bandwidth

Estimation, Data Grid, INCITE Base:Network Monitoring, Data Grid, Transport Protocols

Milestones/Dates/Status Infrastructure development Mon/Yr DONE - develop simple window tuning tool 08/01 08/01 - initial infrastructure developed 12/01 12/01 - infrastructure installed at one site 01/02 01/02 - improve and extend infrastructure 06/02 - deploy at 2nd site 08/02 - evaluate GIMI/DMF alternatives 10/02 - extend deployment to PPDG sites 03/03• Develop analysis/reporting tools - first version for standard apps 02/02 Integrate new apps &net tools - GridFTP and demo 05/05 - INCITE tools 08/02 - BW measure tools (e.g. pathload) 01/03 • Compare & validate tools - GridFTP 09/02 - BW tools 04/03

PingER novel ideas Low impact network performance measurements to most of the Internet connected world providing delays, loss and connectivity information over long time periods Network AND application high throughput performance measurements allowing comparisons, identification of bottlenecks Continuous, robust, measurement, analysis and web based reporting of results available world wide Simple infrastructure enabling rapid deployment, locating within an application host, and local site management to avoid security issues

PingER: Active End-to-end performance monitoring for the Research and Education

communities Tasks: -develop/deploy simple, robust ssh based active end-to-end measurement and management infrastructure -develop analysis/reporting tools -integrate new application and network measurement tools into the infrastructure -compare & validate various tools, and determine regions of applicability www-iepm.slac.stanford.edu




Stability Modeling and Control of Transport Protocols for High-Speed Data GridsNageswara S. Rao, Oak Ridge National Laboratory


Provides controlled end-to-end dynamics for grids over wide-area networks – significant step beyond state-of-the art

Fundamentally new classes of transport methods based on sound analysis and experimentation – inexpensive and easy to use

Provides the needed quality of service for control over wide-area networks for data and instrument grids

CONNECTIONS: Net100 project: will use the proposed instruments and will

provide certain measurement modules Terascale Supernova Initiative can significantly benefit from

the proposed control methods – we are in communication

Milestones/Dates/Status Detailed rigorous analysis: - attractor analysis Feb 02/Feb 03 - conditions of chaos Apr 02/Apr03 Grid network instrumentation design: - sufficiency proofs of measurements Mar 02/ Mar03 - detailed module design June 02 Proof of concept implementations: - high throughput July 02 - bounded higher order delay moments Aug 02/Sept 03 Application and testing: - identification of representative problem Feb 03 - performance study Sept 03

The Novel IdeasDetailed analysis of transport dynamics using non-linear control and chaos theory – showed that TCP generates “complicated” phase space attractorsDeveloped the concept of grid network instruments to perform measurement and traffic engineering using light-weight in-situ modules – analytically showed their performance optimalityNovel transport control methods for end-to-end control for

- high throughput using concurrent window and graded control- controlled dynamics using multiple throttle methods

Understand and Control the End-to-End Transport Dynamics of High-Speed Grids • Detailed analysis of transport processes

• rigorous treatment using non-linear control and chaos theory

• Develop provably effective transport methods for:

•high throughput, and •end-to-end dynamics control

• Implement and test on grid environments



MICS Program Manager: Thomas D. Ndousse

Pushing the Network Simulation EnvelopeW. R. Wing - Oak Ridge National Laboratory

Impact and Connections IMPACT: SSFnet will be the first network simulator able to:

Fully model SciDAC Terascale applications Allow SciDAC developers to tune their applications to

evolving mixed-technology network environments Allow testing/confirmation of future SciDAC-developed

network protocols CONNECTIONS: A key element of SSFnet’s verifiability is our

plan to directly incorporate the Net100/Web100 MIB in the simulator. Comparison of real-life MIB measurements with the SSF-instrumented MIB will provide confirmation of SSFnet simulation fidelity. However, this does require deployment of at least some SciDAC applications on Web100/Net100 platforms


Proposed Milestone Proposed Date Actual Date

Verify Shared-mem architectures - IBM, Compaq, Solaris Q1 - FY02 CompleteDevelop initial DM scheduler Q3 - FY02Develop MIB instrumentation Q4 - FY02Develop application-level IDE Q2 - FY03Develop 2nd-Gen DML-based Scheduler Q4 - FY03Distribute to DOE community Q4 - FY03

SSFnet Novel Ideas

• SSFnet will be the first network simulator with verifiable instrumentation - We plan to include (not model) the Net100/Web100 MIB - Net100/Web100 MIB data will be accumulated for direct comparison• SSFnet will be the first production quality Distributed Memory simulator - Domain Modeling Language will automate decomposition • SSFnet will be the first simulator able to tackle SciDAC-scale problems

SSFnet - Creating a Terascale network simulator that can model SciDAC applications

Tasks:- Verify SM SSFnet on candidate architectures- Develop initial DM version of SSFnet- Develop and verify instrumentation- Develop application-level IDE- Distribute to DOE network research community- Develop 2nd-Gen DM scheduler and DML

Date Prepared 01/08/ 02


MICS Program Manager: T. Ndousse

High-Performance Transport ProtocolsPI: Wu-chun (Wu) Feng, Los Alamos National Laboratory and The Ohio State University


Dynamic Right-Sizing- Auto-tuned, order-of-magnitude increase in throughput.- Vendor adoption, e.g., IRIX, Linux (still in the works)- Potential integration into GridFTP, Web100, Net100.

RAPID- Sliding reliability semantics may result in adoption of

RAPID by LANL large-data visualization team. CONNECTIONS.

Dynamic Right-Sizing: Web100, Net100, DOE Science Grid, Particle Physics Data Grid, Earth System Grid II,

RAPID: The LANL large-data visualization team, previously sponsored by the DOE NGI Corridor One project. Others?


Mon Yr DONE Simulation: Flow-Control Adaptation with Dynamic Right-Sizing

-Protocol Analysis & Design (ns-2) 12/01 12/01-Protocol Testing & Evaluation (rudimentary) 03/02 beta testing

Implementation: Flow-Control Adaptation with Dynamic Right-Sizing -Kernel Space, Linux 2.4.x 07/02 beta testing -User Space, drsFTP 01/03 alpha testing

-Protocol Testing & Evaluation (rudimentary) 03/03-Potential Integration with GridFTP 04/03-Deployment (kernel- & user space) 07/03

Simulation: RAPID-Effect of packet spacing 03/02preliminaries-Definition of API to middleware 03/02preliminaries-Sliding reliability 07/03

The Novel Ideas Dynamic Right-Sizing: TCP Flow-Control Adaptation

for Grids & the Next-Generation Internet Automatically enhance network performance over the

WAN by as much as an order of magnitude while abiding by TCP semantics.

RAPID: Rate-Adjusting Protocol for Internet Delivery Provide smoother QoS support over the best-effort

Internet for grids and NGI while minimizing the need for widespread deployment of DiffServ or IntServ.

Goal: To significantly improve network performance in support of all computing environments, particularly grids and NGI. • TCP/IP Make the network fast but TCP friendly.

• Eliminate TCP’s flow-control bottleneck by automatically tuning buffer sizes.

• RAPID Make the network more adaptable.• Smooth QoS support over a best-effort

network.• User-settable reliability, providing a spectrum

of QoS from unreliable UDP to reliable TCP.

January 16, 2002


MICS Program Manager: Thomas Ndousse-Fetter

Office of Science MICS Division Department of Energy High-Performance Networking Research Program...

Documents

Transcript of Office of Science MICS Division Department of Energy High-Performance Networking Research Program...