MASTS and 46PaQ Lionel Sacks <lsacks@ee.ucl.ac.uk> Saleem Bhatti <s.bhatti@cs.ucl.ac.uk> September 2004 UKLight Town Meeting - MASTS and 46PaQ 1 Technolog y Innovation Network Science MASTS & 46PaQ: Overview Towards Next Gen Networks MASTS Analysis at All Scales Through Time & Space Data Acquisition & Archiving Real-time Monitoring Front End: Compression (Model Free & Model Dependant) MD k2 CD c i MD k2 CD c i Waring 46PaQ Protocol Innovation Behaviour And Performance For QoS and Control Man Data Analysis On Grid Services Front End Enhancements: Advanced Protocols MD k2 CAM MD k2 CD c i ULCC SuperJanet September 2004 UKLight Town Meeting - MASTS and 46PaQ MD k2 CERN Italy 2 Technolog y Innovation Network Science MASTS & 46PaQ: Goals Engineering Terms For NGN MASTS Analysis at All Scales Through Time & Space Towards Next Gen Networks Management & Signalling Time-scales 46PaQ Protocol Innovation Behaviour And Performance For QoS and Control Innovation in QoS & Control Data Acquisition & Archiving Real-time Monitoring Data Analysis On Grid Services Eyes and Ears for Generic technology for high-rate, Front End: UKLight real-time input to Grid Compression (Model Free & Model Dependant) Front End Enhancements: Advanced Protocols MD k2 CD c i MD k2 CD c i Waring Man MD k2 CAM MD k2 CD c i ULCC SuperJanet September 2004 UKLight Town Meeting - MASTS and 46PaQ MD k2 CERN Italy 3 Service to the Community Provide performance information to experiments: “eyes and ears” Provide raw and analysed data in Grid environment Working closely with UKERNA: direct feedback to the academic network infrastructure SuperJANET5 and beyond … Management Info (real time) PoP September 2004 Grid Probe Ana Ana Ana Ana Remote Server Grid Services Stats & analysis Grid Archives UKLight Town Meeting - MASTS and 46PaQ 4 UKLight: MASTS Measurement at All Scales in Time and Space Science Case Functional Objectives Technology Case Output L. Sacks (EE) S. Bhatti (CS) September 2004 D. Parish (EE) I. Philips (CS) Official Name: MATS Monitoring at many scales – in time and space Status: Kick-off September ‘04 ~£1.1M I. Pratt (CL) A. Moore (CL) R. Gibbens (CL) UKLight Town Meeting - MASTS and 46PaQ endace 5 Science Case The ‘Engineering’ of networks has become in part, an empirical science: Complex mixture of: Classical models do not always hold: Applications, Economics, User Behaviour, Technology Evolution Poisson models not applicable for traffic modelling Central Limit Theorem based statistics Almost model free observations required New phenomenology needs to be explored Classic Communications dichotomies chainging: September 2004 High bandwidth, error tolerant (TV, DTV, bulk POTs) Low bandwidth, reliable (web, email, telnet …) UKLight Town Meeting - MASTS and 46PaQ 6 Science Case: Time Traffic Flow Statistics: Unique mix of demands: A never ending source of amusement! Data copying, Interactive / visualisation Control, Video, instrumentation Applications still in development What does this Really look like? The War of the Paradigms: Circuit Switched: Packet Switched: Sporadic transmission, high signaling cost Hybrid: September 2004 Sustained flows, low signaling cost the best of times, the worst of times UKLight Town Meeting - MASTS and 46PaQ 7 Science Case: Space Network Topologies Persistence of flows A never ending source of amusement! Where in the network? Between who? Scaling of capacity: September 2004 From core, through to access UKLight Town Meeting - MASTS and 46PaQ 8 Network Engineering Case For example, Switched Optical Networks: Time distribution of flows – from core edge And just what do we mean by core & edge? Aggregation / Capacity management Switch dimensioning (optical & electronic) UKLight presents an opportunity to measure the evolution of a network for a new set of applications & services. September 2004 UKLight Town Meeting - MASTS and 46PaQ 9 Technology Case Short Time scales: Monitoring up to OC192 is just hard Key application trail of GridProbe Could generate ~ 20TBytes / day Explore real-time Compression Model free & Model Dependant Engineering Challenges Classic Management jobs: Acquire & Archive VPN configuration And correlate with measurements, applications etc. Archive data Serve data & analysis to users through Grid Services Networked Systems Analysis as an e-Science! September 2004 UKLight Town Meeting - MASTS and 46PaQ 10 Example Compression Approach Wavelet analysis can achieve 20:1 compression while loosing ~ 3% critical information DWT would require some fancy FPGA work … September 2004 UKLight Town Meeting - MASTS and 46PaQ 11 UKLight: 46PaQ IPv4 and IPv6 Performance and QoS IPv4 and IPv6 QoS control New protocols Network control and behaviour S. Bhatti (CS) P. Kirstein (CS) September 2004 Status: UKERNA I. Pratt (CL) A. Moore (CL) UKLight Town Meeting - MASTS and 46PaQ Kick-off September ‘04 ~£650K Cisco Sun 12 IPv4 and IPv6 IPv4: Some newer applications benefit from IPv6: still has a long-life ahead! NATs, proxies, etc. auto-config capability addressing and routing mobility and mutli-homing support IPv4 and IPv6 need to be used simultaneously on the same network September 2004 UKLight Town Meeting - MASTS and 46PaQ 13 QoS control Reservations for network flows: “Usual” applications, e.g.: video Research computing applications, e.g.: protected capacity for certain flows large file transfers file-transfers in a distributed pipeline protected capacity for remote-visualisation distributed computational steering Initial experience already with the GRS project: September 2004 http://www.cs.ucl.ac.uk/research/nrs/ UKLight Town Meeting - MASTS and 46PaQ 14 Newer protocols Explicit Congestion Notification (ECN) New protocols: Stream Control Transport Protocol (RFC2960) Datagram Congestion Control Protocol High-performance paths (large BDP paths) Lots retro-fitted into IPv4 but native in IPv6: September 2004 How does this change the network behaviour? UKLight Town Meeting - MASTS and 46PaQ 15 So what happens if … ? How do we use all these things together? Feedback to the development process: at high-speed? with both IPv4 and IPv6? and undertake a protocol performance analysis? networks protocols end-systems How does the network behave? Is it stable and controllable? September 2004 UKLight Town Meeting - MASTS and 46PaQ 16 Expected outcomes: output Experiments with network performance analysis at high-speeds: Feedback to applications: methods metrics tools performance-based adaptation for applications Network engineering insights September 2004 UKLight Town Meeting - MASTS and 46PaQ 17 Expected outcomes: questions How do we design and engineer high-speed protocols and networked systems that: make efficient use of QoS and congestion control? can provide feedback to applications? How can applications react in a timely manner to feedback from the network? How do we design and build high-speed data acquisition and processing systems? September 2004 real-time requirements? UKLight Town Meeting - MASTS and 46PaQ 18 View points in space and time User plane: Control plane: user-to-network network-to-network Management plane: user protocols – data transfer management plane control plane user plane (data) network performance network monitoring Timescales: September 2004 sub-seconds to years UKLight Town Meeting - MASTS and 46PaQ 19 UKLight - First Observation! Economics & Technology Evolution Original Tender: 2.5Gigbit/s or 10Gigbit/s At the time we saw the tenders: No price difference! But 1x10Gigbit/s is really interesting: Especially if not-aggregated Dynamic network landscape! September 2004 UKLight Town Meeting - MASTS and 46PaQ 20 Contact Details MASTS: http://www.ee.ucl.ac.uk/~lsacks/ Will move to http://www.masts.uklight.ac.uk/ 46PaQ: September 2004 http://www.cs.ucl.ac.uk/staff/S.Bhatti/46paq/ Will move to http://www.46paq.uklight.ac.uk/ UKLight Town Meeting - MASTS and 46PaQ 21