GRID.IT GRID.IT Programma Nazionale della Ricerca 2001-03 Bando FI RB Progetto Strategico “Tecnologie Abilitanti per la Società della Conoscenza” Progetto obiettivo “Reti e Netputing” Progetto “Piattaforme abilitanti per griglie computazionali a elevate prestazioni orientate a organizzazioni virtuali scalabili” GRI D.I T (Resp.: Prof. Marco Vanneschi, UniPi/ CNR) CHECK POI NT – 1° anno WP1 – GRI D ORI ENTED OPTI CAL SWI TCHI NG PARADI GMS Resp. Piero Castoldi Pisa, 6-8 Ottobre 2003 Relazione con l’UdR CNIT Resp. UdR CNIT: Prof. Giancarlo Prati Articolazione su due workpackages • WP 1 – Grid oriented optical switching paradigms (Resp. Piero Castoldi) • WP 2 - High-performance photonic test-bed (Resp. Stefano Giordano) GRID.IT GRID.IT 2
Specifications on interaction among areas Area 4: Applications WP10 WP11 WP12 WP13 WP14 WP7 WP8 WP9 WP4 WP6 Area 1: Middleware and Programming Tools A1.1, A1.2, A1.3, A1.4, A1.5 WP1 WP2 WP3 WP5 Area 3: Grid deployment Area 2: Photonic testbed A1.1, A1.2, A1.4 GRID.IT GRID.IT 3 Istituzioni che partecipano al al WP1 • Laboratorio Nazionale di Reti Fotoniche, Pisa Know-how: all 5 major areas of optical networks and photonic technologies (routing and switching, trasmission, amplification, systems) • UdR CNIT at � Università Politecnica delle Marche Know-how: enabling technologies and OXC WXC architectures � Università di Bologna Know-how: architectures for optical packet switching networks � Università di Trento Know-how: traffic models and resource allocation • University of Texas @ Dallas (UTD) Know-how: control plane and reliability for optical networks GRID.IT GRID.IT 4
WP1 breakdown WP 1 - Grid oriented optical switching paradigms (Resp. P. Castoldi) [70% of the total CNIT UdR funding] Technical Board member: Dr. Luca Valcarenghi (assistant professor). • Activity 1 – Connections, topologies and network service models 8% 1° year 2° year (Resp. R. Battiti) [Lab PI , TN] • Activity 2 – Grid computing on state-of-the-art optical networks 15% 1° year 2° year (Resp. P. Castoldi) [Lab PI , TN, UTD] • Activity 3 – Migration scenarios to intelligent flexible optical networks (Resp. F. Callegati) 15% [Lab PI , BO, TN, UTD] 1° year 2° year • Activity 4 – Control plane and network emulation for optical packet switching networks 15% (Resp. A. Fumagalli) [Lab PI , TN, UTD] 2° year 3° year • Activity 5 – Enabling technologies for optical switching networks 17% (Resp. G. Cancellieri) [Lab PI , BO, AN] 1° year 2° year 3° year GRID.IT GRID.IT 5 Optical Grid Networking Roadmap Virtual Processor (VP) f(x1,d) connection (d1,dn) f(x2,d1) f(x,d) (d2) Emitter (E)/ Collector (C) Distributed Virtual f(xn,dn) Shared Memory (DVSM) GRID.IT GRID.IT 6
Details about activity 1 Activity 1 – Connections, topologies and network service models 8% 1° year 2° year (Resp. R. Battiti) [Lab PI, TN , UTD] f(x1,d) f(x1,d) MILESTONES and DELIVERABLES (1-18 months) (d1,dn) (d1,dn) f(x2,d1) f(x2,d1) � Traffic models to represent computer generated loads f(x,d) f(x,d) [technical report] (d2) (d2) � Resource Allocation Architecture (RAA) to map network f(xn,dn) f(xn,dn) logical topology requests from GRID to the optical network and virtual topology definition (VTD) based on given traffic f(x1,d) f(x1,d) patterns [technical report and software tool] (d1,dn) (d1,dn) f(x2,d1) f(x2,d1) f(x,d) f(x,d) Added value for GRI D.I T project: (d2) (d2) � Use of traffic models for performance evaluation / f(xn,dn) f(xn,dn) capacity planning (input to other activities) � Efficient traffic engineering rules for state of the art grid-oriented networks GRID.IT GRID.IT 7 Achieved Results in Traffic Models (1) � Development of a WinPCap-based tool for traffic analysis (for comparisons between models and actual traffic). Ddum p is a useful tool for traffic capture and analysis and it provides several functionalities: � Compatibility with libpcap and interoperability with Ethereal � Tree and database packet representation � Packet arrival / interarrival time distribution graphs � Distribution matching with traffic models (Poisson, binomial) � Possibility of packet header modification / trace file editing � Data export to libpcap / ns-2-compatible files � Traffic generation � Network status evaluation (ping, traceroute) � Written in MS Visual C+ + w/ WinPcap libraries GRID.IT GRID.IT 8
Achieved Results in Traffic Models (2) An open-source tool for traffic measurement and analysis, called Ddump, available online http://dit.unitn.it/ ∼ ddump GRID.IT GRID.IT 9 Achieved Results in Traffic Models (3) Validation of models for complete TCP characterization (closed form) by simulation Single TCP source λ 1 µ 1 � Single source m odel : M/ P/ 1/ cwnd queue, F 2 i.e. Poisson arrivals (M), TCP parametrized T w,1 service distribution (P), 1 server, cwnd= 64 λ 2 µ 2 Bottleneck packets, outgoing link 1 Mbps, packets of 1024 F 2 λ µ bits Σ T w,2 � Bottleneck ( router) m odel: modeled as B M/ M/ 1/ B, i.e. with Poisson arrivals and service λ N µ N time, 1 server, buffer capacity of B= 128 F 2 packets, outgoing link of 10 Mbps T w,N 9000 10000 Sim 8000 Model 8000 7000 6000 6000 5000 4000 4000 Sim 3000 2000 2000 Model 1000 0 0 0,001 0,005 0,01 0,5 1 1 5 10 15 20 30 Average throughput (Mbps) for different ACK delay Average throughput using NS-2 simulator (“Sim”) and the values using NS-2 simulator (“Sim”) and the proposed model (“Model”) for different numbers of TCP proposed model (“Model”). sources and a fixed bottleneck delay. 10 TCP sources transmitting at 1Mbps are considered. GRID.IT GRID.IT 10
Achieved results on RAA and VTD (1) 1) FID ( x ), First Improve Dynamic load balancing 1. Route the LSPs using some routing scheme (CBR) 2. Repeat for MAXITER iterations: a. Find the most congested edges (MCE) in the network (residual capacity less than x ) b. For each most congested edge: � try to re-route only one of the LSPs on the MCE over another edge � IF congestion is reduced, accept move, and ask the ingress LSR to re-route this LSP 2) LFID, Lazy First-Improve Dynamic Load balancing � re-route a single LSP according to FID only if the new one cannot be accommodated GRID.IT GRID.IT 11 Achieved Results in RAA and VTD (2) Comparison with existing preventive RAA algorithms (MHA, i.e. minimum hop routing and MIRA, i.e. minimum interference routing). Parameters: • Arrival rate with rate λ • Duration of the transactions 1/ µ • Threshold for FID operation 0.01 • 12 units of capacity in the light link • 48 units of capacity in the dark links • LSP bandwidth demand in between 0.01 and 0.04 units of capacity Number of rejected LSPs vs. network load GRID.IT GRID.IT 12
Publications on activity 1 of WP1 1. D. Agrawal, F. Granelli, A Queuing Model for Steady-State Behaviour of TCP in Performance Evaluation of Telecommunication Networks , Softcom 2003, Oct. 2003. 2. M. Brunato, R. Battiti, E. Salvadori, Load Balancing in WDM Networks through Adaptive Routing Table Changes . In Networking, number 2345 in Lecture Notes in Computer Science, pages 289–301, Pisa - Italy, May 2002. Springer Verlag. 3. M. Brunato, R. Battiti, E. Salvadori. Dynamic Load Balancing in WDM Networks . Optical Networks Magazine, September 2003. In press. 4. E. Salvadori, R. Battiti. A Load Balancing Scheme for Congestion Control in MPLS Networks . Accepted in IEEE Symposium on Computers and Communications – ISCC 2003, Antalya - Turkey. 5. E. Salvadori, R. Battiti, F. Ardito. Lazy Rerouting for MPLS Traffic Engineering. Submitted to IEEE Globecom 2003. + 4 DIT Technical Reports GRID.IT GRID.IT 13 Details about activity 2 Activity 2 – Grid computing on state-of-the-art optical networks 15% (Resp. P. Castoldi) [ Lab PI , TN, UTD] 1° year 2° year MILESTONE and DELIVERABLES (1-12 months) � Static and dynamic routing and wavelength assignment (RWA) of optical connections [technical report and software tools] MILESTONE and DELIVERABLES (6-18 months) � Reliability schemes exploiting traffic granularity (GMPLS) for protection and restoration [technical report and laboratory testbed] � GMPLS based control plane for dynamic distributed resource allocation [technical report and laboratory testbed] GRID.IT GRID.IT 14
Routing and Wavelength Assignment (RWA) f(x1,d) f(x1,d) (d1,dn) (d1,dn) f(x2,d1) f(x2,d1) f(x,d) f(x,d) (d2) (d2) f(xn,dn) f(xn,dn) What has been promised: • Blocking probability of static and dynamic RWA schemes • Assessment of benefits of wavelength selection and/or wavelength conversion • Reduction of complexity by sparse or limited wavelength selection and conversion Advantages to GRID.IT • Guidelines for optical network testbed implementations based on a circuit switching paradigm • Complexity requirements for the nodes GRID.IT GRID.IT 15 RWA - Considered Node architectures (1) • No wavelength selection • No wavelength conversion Present situation • Wavelength selection • No wavelength conversion GRID.IT GRID.IT 16
RWA – Considered Node architectures (2) • No wavelength selection • Wavelength conversion • Wavelength selection • Wavelength conversion Best performance Most expensive GRID.IT GRID.IT 17 Simulation results – Blocking probability NOWS WC NOWS WC WS NOWC WS NOWC GRID.IT GRID.IT 18
Recommend
More recommend
