UKLIGHT: projects at UCL Saleem Bhatti - PowerPoint PPT Presentation

UKLIGHT: projects at UCL Saleem Bhatti http://www.cs.ucl.ac.uk/staff/S.Bhatti/ 1

Challenges • High-speed (multi-Gb/s) operation:  network engineering for high-speed • QoS at high speed within the core network • QoS for individual application flows:  state management and site-to-site signalling • Operation of protocols:  for example TCP congestion control proposals 2

Background work MB-NG http://www.mb-ng.net/ Managed Bandwidth Next Generation ~£900K, Co-I GRS http://www.cs.ucl.ac.uk/research/nrs/ Grid Resource Scheduling ~£138K, PI 3

GRS and MB-NG • GRS:  site-to-site per-flow QoS reservations  decentralised  application-level APIs  (use MB-NG network as test-bed) • MB-NG:  SuperJANET4 dev net  DIFFSERV across core  high-speed TCP experiments  (dynamic QoS-control using GRS) 4

from http://www.ja.net/ 5

Experiments: SuperJANET4 Warrington SuperJANET4 Development Network (from the MB-NG project) Manchester RAL ULCC core router - Cisco 12xxx UCL edge router - Cisco 7609 2.5Gb/s POS 6

Activities • QoS performance analysis:  DIFFSERV EF • QoS control software (NRS):  released in May 2004 (open source) http://www.cs.ucl.ac.uk/research/nrs/  dynamic, on-demand, site-to-site QoS control  modular system - other sub-net technologies  (dynamic QoS control recorded demo on web site) • TCP performance analysis:  multiple TCP implementations  exploration of practical usage  end-system limitations 7

The future - questions raised • What happens with a more heterogenous traffic mix? • How does the system scale?  number of flows?  amount of traffic?  size of network?  heterogeneity of traffic? • What is a sensible profile of mechanisms to use?  congestion control? (HS-TCP? Scalable-TCP? ECN? XCP? etc.)  queue management?  QoS control? • Is TCP (+ mods) actually useful for Gb/s?  some other, new transport protocol perhaps? e.g. DCCP? • How should the application get involved? 8

UKLIGHT • The need for a high-speed network facility for research :  transmission/photonics and switching  networking  applications • Link to other existing facilities world-wide:  Global Lambda Infrastructure Facility (GLIF)  http://www.glif.is/ UKLIGHT http://www.uklight.ac.uk/ ~£6.5M, Editor of the Scientific Case, member of the UKLIGHT TAG many contributors - http://www.cs.ucl.ac.uk/staff/S.Bhatti/uklight-proposal/ 9

UKLIGHT - GLIF from http://www.glif.is/ 10

Current work - just starting 46PaQ http://www.cs.ucl.ac.uk/research/46paq/ ~£650K, PI: IPv4 & IPv6 Performance and QoS MASTS http://www.ee.ucl.ac.uk/~lsacks/acse/masts/ ~£1.2M, Co-I: Measurements at All Scales in Time and Space ESLEA Exploiting Switched Lightpaths for E-science Applications MRC ~£1.1M, Co-I 11

Project links ESLEA MB-NG MASTS 46PaQ (core, UK) UKLIGHT (international) GRS (edge) StarLight NetherLight (Chicago) (Amsterdam) GRS - http://www.cs.ucl.ac.uk/research/nrs/ 46PaQ - http://www.cs.ucl.ac.uk/research/46paq/ MB-NG - http://www.mb-ng.net/ MASTS - http://www.ee.ucl.ac.uk/~lsacks/acse/masts/ UKLIGHT - http://www.uklight.ac.uk/ ESLEA - URL TBA 12

“Plan A” : a work-in-progress Applications ESLEA Applications and the Network Towards Network Science Next 46PaQ MASTS Generation Protocol Innovation Analysis at All Scales Behaviour And Performance Networks in Time & Space For QoS and Control Technology Innovation Data Acquisition, Data Analysis and Real-time Monitoring Storage & Archiving Traffic Modelling Compression (Model Free & Model Dependant) Front End Enhancements: Advanced Protocols SuperJANET and UKLIGHT 13

Potential output • High-speed experiments & performance analysis:  methods, metrics, tools • Better (performing) implementations:  end-systems design and engineering • Feedback to applications:  performance-based adaptation for applications • Traffic models:  network management, planning and design • Network engineering insights • Services to the community:  researchers, operational and applications users 14

Other questions forming … • How do we design and engineer high-speed protocols and networked systems that:  make efficient use of QoS and congestion control?  have appropriate instrumentation in the network?  can provide feedback to applications? • How can applications react in a timely manner to feedback from the network?  change the design of applications?  new transport protocols and/or systems APIs?  instrumentation and engineering of the end-systems? • Watch this space ….  15

General areas of work Practical work in three broad areas: • 1. networking in the large 2. scalability and evolution of networked systems 3. high-performance networked systems 4. highly-integrated networked systems Some common key threads: •  control, management and evolution of networked systems interactions: end-system, network, application and user   real experiments on real networks HEN - Heterogeneous Experimental Network: •  “lights-out”, L2/L3-configurable network lab  work-in-progress 16